The Theory of Turbulence - Subrahmanyan Chandrasekhar's - Spiegel E.A. (Ed.)

Lecture Notes in Physics
Founding Editors: W. Beiglböck, J. Ehlers, K. Hepp, H. Weidenmüller

Editorial Board
R. Beig, Vienna, Austria
W. Beiglböck, Heidelberg, Germany
W. Domcke, Garching, Germany
B.-G. Englert, Singapore, Singapore
U. Frisch, Nice, France
F. Guinea, Madrid, Spain
P. Hänggi, Augsburg, Germany
W. Hillebrandt, Garching, Germany
R.L. Jaffe, Cambridge, MA, USA
W. Janke, Leipzig, Germany
R.A.L. Jones, Sheffield, UK
H.V. Löhneysen, Karlsruhe, Germany
M. Mangano, Geneva, Switzerland
J.-M. Raimond, Paris, France
M. Salmhofer, Heidelberg, Germany
D. Sornette, Zurich, Switzerland
S. Theisen, Potsdam, Germany
D. Vollhardt, Augsburg, Germany
W. Weise, Garching, Germany
J. Zittartz, Köln, Germany
The Lecture Notes in Physics
The series Lecture Notes in Physics (LNP), founded in 1969, reports new develop-
ments in physics research and teaching – quickly and informally, but with a high
quality and the explicit aim to summarize and communicate current knowledge in
an accessible way. Books published in this series are conceived as bridging mate-
rial between advanced graduate textbooks and the forefront of research and to serve
three purposes:
• to be a compact and modern up-to-date source of reference on a well-defined
topic
• to serve as an accessible introduction to the field to postgraduate students and
nonspecialist researchers from related areas
• to be a source of advanced teaching material for specialized seminars, courses
and schools
Both monographs and multi-author volumes will be considered for publication.
Edited volumes should, however, consist of a very limited number of contributions
only. Proceedings will not be considered for LNP.
Volumes published in LNP are disseminated both in print and in electronic for-
mats, the electronic archive being available at springerlink.com. The series content
is indexed, abstracted and referenced by many abstracting and information services,
bibliographic networks, subscription agencies, library networks, and consortia.
Proposals should be sent to a member of the Editorial Board, or directly to the
managing editor at Springer:
Christian Caron
Springer Heidelberg
Physics Editorial Department I
Tiergartenstrasse 17
69121 Heidelberg/Germany
christian.caron@springer.com
For other titles published in this series, go to

www.springer.com/series/5304
Edward A. Spiegel
Editor
The Theory
of Turbulence
Subrahmanyan Chandrasekhar’s
1954 Lectures
Notes prepared by E.A. Spiegel

Editor
Dr. Edward A. Spiegel
Dept. Physics, Pupin Laboratory
Columbia University
West 120th St. 538
10027 New York, NY
USA
eas5@columbia.edu
Spiegel, E.A., The Theory of Turbulence: Subrahmanyan Chandrasekhar’s 1954

Lectures, Lect. Notes Phys. 810 (Springer, Dordrecht 2011), DOI 10.1007/978-94-
007-0117-5
ISSN 0075-8450 e-ISSN 1616-6361

ISBN 978-94-007-0116-8 e-ISBN 978-94-007-0117-5
DOI 10.1007/978-94-007-0117-5
Springer Dordrecht Heidelberg London New York
Library of Congress Control Number: 2010938378
© Springer Science+Business Media B.V. 2011

No part of this work may be reproduced, stored in a retrieval system, or transmitted in any form or by
any means, electronic, mechanical, photocopying, microfilming, recording or otherwise, without written
permission from the Publisher, with the exception of any material supplied specifically for the purpose
of being entered and executed on a computer system, for exclusive use by the purchaser of the work.
Cover illustration: eStudio Calamar S.L.
Printed on acid-free paper
Springer is part of Springer Science+Business Media (www.springer.com)

DE GUICHE.
Car, lorsqu’on les attaque, il arrivent souvent . . .
Qu’un moulinet de leurs grands bras chargés de toiles
Vous lance dans la boue !. . .
CYRANO.
Ou bien dans les étoiles !
Prologue
At a Midwest Neighborhood Astronomy Meeting in the spring of 1953, I gave a

fairly naive ten-minute talk about observations of turbulence in a supergiant star.
Chandrasekhar was sitting near the front so, of course, I was nervous. After all, I
had first heard of fluid dynamical turbulence as an undergraduate just three years
before that in a lecture given by Chandrasekhar himself, when he looked about like
the photo shown here from the Publications of the Astronomical Society of the Pa-
cific. Hearing that talk was a bit like the butterfly effect that is a popular mantra
among students of chaos theory. When I had announced to one of my professors
that I wanted to pursue theoretical rather than observational research, he asked what
I wanted to work on. I quickly responded with the only explicit theoretical topic
whose name I knew: turbulence. And so I had sealed my fate on the basis of a rec-
ollection that brought me to my present uncomfortable situation.
Half way into my talk, Chandra, as he shall be known in these notes, blew his
nose rather loudly and this caused my knees almost to buckle. But I got through it all
and was invited to have lunch with him afterwards. In the course of that meal, Chan-
dra invited me to come and study with him at the Yerkes Observatory in Williams
Bay Wisconsin. We compromised on a visit for the following summer. This was in
1954 when I was at the end of my second year as an astronomy graduate student in
the University of Michigan and had not had the benefit of any real graduate courses
in physics or mathematics. However, I had been studying fluid dynamics on my own
and that was perhaps the reason Chandra invited me. He had been at work on fluid
dynamics, especially on convective instability, and was homing in on the problem
of turbulence.
A few years before that, Bengt Stromgren, the then director of the Yerkes Ob-
servatory, had written that convection was the single most important problem in the
theory of stellar atmospheres, and stellar atmospheres was then perhaps the most
dynamic subtopic on the frontier of astrophysics. On the other hand, the way stellar
convection was being treated at the time was such as to make any theoretician with
a conscience feel the need to do something to improve the situation. Since convec-
tion in stars like the sun is turbulent by the measures that are used to make that
judgement, it seemed that the first order of business was to understand turbulence.
vii
viii Prologue
Chandra had already begun preparing himself to do this some years earlier. The 1949
Astrophysical Journal contains his “Turbulence: A Physical Theory of Astrophysi-
cal Interest,” which was an account of his Russell Lecture, a major annual lecture of
the American Astronomical Society. And so, in the summer of 1954, Chandra had
set himself the goal of tilting at turbulence. These notes are an account of the series
of lectures on the subject he presented at the Observatory that summer.
In order to support myself during my stay at the Observatory, I worked as the
research assistant of W.W. Morgan, one of the great observational astronomers of
that period. Among other assignments, I was expected to take spectra for Morgan on
the 40 Yerkes refracting telescope. Though I am a klutz and never had any practical
skill as an observer, I found this work exhilarating since I was taking spectra of Be
stars, hot stars with emission lines, and I found them quite interesting (and still do).
Moreover, Morgan liked to teach me things and that was also a great pleasure for
me. He was a naturalist and could look at a picture of the astronomical sky and
observe things that ordinary mortals did not. He would have me sit down before an
astronomical microscope and have me look into it. “What do you see in there?” he
would ask. And I would tell him that I saw this or that. To which he would replay,
“Don’t you see the XXX?” So I would look till I finally saw the desired object (or
thought I did). From such sessions, I came to appreciate an aspect of astronomical
discovery that few people are aware of. Great observers really do observe. Like the
eye of a good artist who observes things in heir1 surroundings that many people do
not notice, there is also a naturalist’s eye which is a gift that Morgan had and that he
used to great effect.
At that time, Yerkes Observatory was a great astronomical center with a staff
of renowned astronomers. Heady encounters were a daily occurrence for me from
the day I arrived from northern Michigan where I had gone to witness a total solar
eclipse. I had presented myself to Barbara Perkins, the main secretary of the ob-
servatory, and stood there waiting for her to decide what to do with me when the
Director, Bengt Stromgren, came by. He was a most impressive figure who (like
Karl Schwarzschild) had written his first paper in his teens. Barbara introduced him
to the “newly arrived graduate student” and he greeted me warmly, invited me into
his office and asked me to sit by him on a couch with a writing pad between us. He
asked me what I had been working on and, after I finished, he asked me to repeat
my story so that he could be sure he understood it. I was to encounter such kind-
ness from the leading astronomers of the period but I did not then know that this
was normal behavior among them. Still, no one could match Stromgren for gracious
behavior, though Paul Ledoux and Albrecht Unsöld were up there.
Somewhat awed, I returned to Barbara Perkins and to the problem of what to do
with me when a cheerful Britisher came by and told her to put me in his “room.”2
So for a summer I became the office mate of Raymond Hide. Though our biological
1 To avoid such abuses of grammar as substituting ‘their’ for ‘him-or-her’ I simply remove the ‘t’
to create a gender neutral, singular possessive pronoun or adjective. Similarly, I use hey and hem
for the corresponding nominative and accusative pronouns.
2 English for the American “office”.
Prologue ix
ages were not very different, our academic ages were. Raymond was Chandra’s
postdoc and I a mere graduate student (in astronomy at that). The arrangement was
fine for me since I was poised to learn anything and everything I could and Raymond
knew quite a lot even then. The first thing he taught me was drinking. I had left New
York City for California at age seventeen where the legal drinking age was eighteen,
so I did not have a chance there. In California the drinking age was twenty-one but I
left for Michigan just after I turned twenty-one. No one drank at home when I grew
up, so I was a novice and was not initiated into that activity until the end of my
first day in Raymond’s office, when we went for some sherry. That introduction to
social drinking was my beginning and I have continued the practice ever since. I am
reminded of a remark of W.C. Fields who said “A woman drove me to drink and I
never had the courtesy to thank her.” Raymond led me to drink and, like Fields, I
have been remiss in my obligation to express gratitude until now that his eightieth
birthday provides the chance for redress with a paper in his honor.
Another important aspect of my learning experience that summer was that there
seemed to be few colleagues then at Yerkes with whom Chandra could discuss the
problems he was working on so that he often spent time discussing general scientific
matters with me during my stay. This was also part of my extraordinary experience.
Of course, the high point of that experience, represented by this little volume, was
Chandrasekhar’s lecture course on Turbulence.
For me, the first surprise in going to those lectures was the size of the audience.
Though I never counted heads, I estimate from the image in my mind’s eye that there
were more than twenty people in the room and that most of them stayed with the
lectures to their conclusion. This was a large audience for mathematical lectures in
an astronomy department, but the observatory staff was relatively large and it was,
after all, Chandra up in front. I imagine one could go to his lectures, not understand
a word and still be enthralled. In his writing, his lecturing, his conversation, and
even in his handwriting, Chandra had a pronounced style that was somewhat hyp-
notic and invited imitation. But he was inimitable, particularly in his mathematical
pyrotechnics, so that most of the young people who tried to be like him became
mere bad copies. I recognized that danger and (with much embarrassment) declined
his invitation to remain at Yerkes as his student and returned to Michigan to fumble
along on my own after that summer.
I suppose I should apologize for going on in this way but I wanted to give an idea
of the surroundings in which things took place at Yerkes Observatory in that time,
as well as my lack of scientific sophistication. That I was a tabula rasa is important
to understanding the character of these notes.
What was newest to me in Chandra’s turbulence lectures was his deft manipula-
tion of mathematical expressions. The mathematical tone I had met in various books
on fluid dynamics and other branches of physics had none of his virtuosity, so Chan-
dra’s mathematical usage was eye-opening. (I further cured my deficiencies later by
taking a one year course in graduate quantum mechanics given by K.M. Case, who
was said to be the first person to break the four-minute mile—at the blackboard! His
was an entirely different style . . . but I digress.)
After each lecture by Chandra, as I transcribed my notes into a bound note-
book (of a kind popular among physics graduate students in Ann Arbor), I filled
x Prologue
in details of the derivations that Chandra had not included; they would have
been obvious to anyone with a greater knowledge of mathematical physics than
mine. This feature will no doubt make these notes seem very elementary to
many people. On the other hand, I tried to use Chandra’s language as much as I
was able to preserve the spirit of his presentation. I even included a few of his
asides.
The notes as presented here are as I had set them down in 1954 except that I have
proofread them for the first time to remove any obvious typos. Naturally, this far
along, I see things that I myself would rather do differently but that would not be in
the spirit of this venture. Nor have I added references or included an introductory
history of the subject as the referee (Uriel Frisch) proposed. The only attributions in
the notes are those made by Chandra who gave very few references. I have added a
brief epilogue that does have a reference or two. Though I have been at work on a
longer concluding section for some while, Joe Keller convinced me not to delay the
publication of these notes till I had completed that epiproject and, in my experience,
his advice should always be taken. Moreover, 2010 is the centennial of Chandra’s
birth, so there was no other reasonable choice.
A further point needs to be explained. S.K. Trehan once took the notes from
a course given by Chandra at the University of Chicago and published them with
the University of Chicago Press with the title “Plasma Physics” and with the au-
thor given as S. Chandrasekhar. Chandra once told me that he was unhappy about
this as he had not been consulted about the contents of the book. To avoid hav-
ing a sense of responsibility for those contents he had never looked into the book.
This feeling of Chandra posed a dilemma for me that I have solved in my own
way. (Though I am named as editor herein, that is not the “way” I had in mind.)
Still, I believe that anyone who reads these notes will sense Chandra’s presence
behind the naivety of the presentation. I have long agonized over whether to al-
low myself to make slight changes in the presentation given here. But once that
starts it does not end till the whole thing is rewritten and a different book emerges.
So despite the deficiencies of my interpolated intermediate steps of 1954, I have
left things as they were except for a few slight changes of word order for clar-
ity.
Thus, I reluctantly send off these notes. Though there is much more I should do
with them, the time has come and I have only to thank the editorial staff at Springer
for their gentle prodding and Steve Lyle for his fine LaTeXing. And yet there is
another issue on my mind and that concerns the inclusion of the anecdotes or side
remarks with which Chandra would enliven his lectures. Those are not in the notes
but I shall mention two right here.
After Chandra produced his deft solution of Heisenberg’s equation for the turbu-
lent energy described in the notes, he annunciated this ‘theorem’: Given any great
man, he has a weakness through which you can associate yourself to him. Then he
smiled and added “Heisenberg can’t solve differential equations.”
The other story, one that gave him much pleasure to relate, concerned a visit
to Fermi’s office on the Chicago University campus to which Chandra went once
a week to give his courses and to edit the Astrophysical Journal. He had gotten
Prologue xi
interested in turbulence and wondered what Fermi might know of the subject. On
being asked that question, Fermi thought a moment and said he did not know the
subject but he thought it should go something like this. Whereupon he went to the
blackboard and off the cuff gave a derivation of Kolmogorov’s law that Chandra
reproduced in the lectures. I think that Chandra felt that Fermi had had no previous
knowledge of this law and, magician that he was, pulled Kolmogorov’s spectrum
right out of a nonexistent hat.
xii Prologue
S. Chandrasekhar (1910–1995), originally published in 1952 in the publications

of the Astronomical Society of the Pacific, Vol. 64, No. 377, p. 55 and reproduced
with kind permission of this Society © 1952
Prologue xiii
A cascade. Photograph by Antonello Provenzale

Contents
Prologue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
1 The Turbulence Problem . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 The Meaning of ‘Turbulence’ . . . . . . . . . . . . . . . . . . . . 1
1.2 Two Fundamental Aspects of Turbulence . . . . . . . . . . . . . . 2
2 The Net Energy Balance . . . . . . . . . . . . . . . . . . . . . . . . . 3
3 Interchange of Energy Between States of Motion . . . . . . . . . . . 7
4 Some Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
4.1 On the Harmonic Analysis . . . . . . . . . . . . . . . . . . . . . . 11
4.2 On the Concept of Isotropy . . . . . . . . . . . . . . . . . . . . . 11
4.3 On the Possibility of a Universal Theory . . . . . . . . . . . . . . 12
5 The Spectrum of Turbulent Energy . . . . . . . . . . . . . . . . . . . 13

5.1 The Spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
5.2 An Equation for the Spectrum . . . . . . . . . . . . . . . . . . . . 14
6 Some Preliminaries to the Development of a Theory

of Turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
7 Heisenberg’s Theory of Turbulence . . . . . . . . . . . . . . . . . . . 21

7.1 The Fundamental Equation of the Theory . . . . . . . . . . . . . . 21
7.2 Chandrasekhar’s Solution of (7.17) for the Case of Stationary
Turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
8 Other Derivations of the k −5/3 Law . . . . . . . . . . . . . . . . . . . 29

8.1 Fermi’s Approach . . . . . . . . . . . . . . . . . . . . . . . . . . 29
8.2 Kolmogorov’s Theory . . . . . . . . . . . . . . . . . . . . . . . . 30
8.3 The Method of von Neumann . . . . . . . . . . . . . . . . . . . . 31
8.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
xv
xvi Contents
9 An Alternate Approach: Correlations . . . . . . . . . . . . . . . . . 35
10 The Equations of Isotropic Turbulence . . . . . . . . . . . . . . . . . 37

10.1 The Concept of Isotropy . . . . . . . . . . . . . . . . . . . . . . . 37
10.2 Qij as an Isotropic Tensor . . . . . . . . . . . . . . . . . . . . . . 38
10.2.1 Two More Examples . . . . . . . . . . . . . . . . . . . . . 39
10.3 Solenoidal Isotropic Tensors . . . . . . . . . . . . . . . . . . . . . 40
10.3.1 Isotropic Vectors Li . . . . . . . . . . . . . . . . . . . . . 40
10.3.2 Further Manipulations . . . . . . . . . . . . . . . . . . . . 42
10.3.3 The Isotropic Third Order Tensor, Tij k . . . . . . . . . . . 44
11 The Karman-Howarth Equations . . . . . . . . . . . . . . . . . . . . 49
12 The Meanings of the Defining Scalars . . . . . . . . . . . . . . . . . . 51
13 Some Results from the Karman–Howarth Equation . . . . . . . . . . 55

13.1 The Taylor Microscale . . . . . . . . . . . . . . . . . . . . . . . . 55
13.2 The Study of the Decay of Turbulence . . . . . . . . . . . . . . . 56
13.3 The Connection Between the Karman–Howarth Equation
and the Kolmogorov Theory . . . . . . . . . . . . . . . . . . . . . 57
13.3.1 The Double Correlation . . . . . . . . . . . . . . . . . . . 57
13.3.2 The Triple Correlation . . . . . . . . . . . . . . . . . . . . 58
14 The Relation Between the Fourth and Second Order Correlations

When the Velocity Follows a Gaussian Distribution . . . . . . . . . . 61
14.1 Some Properties of the Gaussian Distribution . . . . . . . . . . . . 61
14.1.1 One-Dimensional Gaussian Distribution . . . . . . . . . . 61
14.1.2 n-Dimensional Gaussian Function . . . . . . . . . . . . . 64
14.1.3 Two-Dimensional Gaussian Function . . . . . . . . . . . . 67
14.2 Addition Theorem for Gaussian Distributions . . . . . . . . . . . . 67
14.3 Proof of (14.2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
15 Chandrasekhar’s Theory of Turbulence . . . . . . . . . . . . . . . . 71
16 A More Subjective Approach to the Derivation of Chandrasekhar’s

Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
17 The Dimensionless Form of Chandrasekhar’s Equation . . . . . . . 83
18 Some Aspects and Advantages of the New Theory . . . . . . . . . . . 85

18.1 A Mathematical Justification of the Assumptions
of the Heisenberg Theory . . . . . . . . . . . . . . . . . . . . . . 85
18.2 Compatibility with the Kolmogorov Theory . . . . . . . . . . . . 86
19 The Problem of Introducing the Boundary Conditions . . . . . . . . 89
20 Discussion of the Case of Negligible Inertial Term . . . . . . . . . . . 91

Contents xvii
21 The Case in Which Viscosity Is Neglected . . . . . . . . . . . . . . . 95
22 Solution of the Non-Viscous Case Near r = 0 . . . . . . . . . . . . . . 99
23 Solution of the Heat Equation . . . . . . . . . . . . . . . . . . . . . . 101
24 Solution of the Quasi-Wave Equation . . . . . . . . . . . . . . . . . . 103
25 The Introduction of Boundary Conditions . . . . . . . . . . . . . . . 111
Epilogue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
Chapter 1
The Turbulence Problem
1.1 The Meaning of ‘Turbulence’
In a general way we may say that turbulence is a phenomenon associated with the
velocity state of a fluid medium. It exists in incompressible fluids or with small
(with respect to the velocity of sound) velocities in a compressible fluid, so that
to sufficient accuracy, the problems of turbulence may be discussed in terms of
incompressible fluids.
We know, moreover, that the turbulence phenomenon is described by the more
usual equations which apply to problems of hydrodynamics. In particular, the
Navier–Stokes equation
∂ui ∂ 1 ∂p
+ (ui uj ) = − + ν∇ 2 ui , (1.1)
∂t ∂xj ρ ∂xi
and the equation of continuity for an incompressible fluid
∂ui
= 0, (1.2)
∂xi
apply to the turbulence phenomenon. Here, the symbols all carry their usual mean-
ing, and (as in the remainder of these notes) the summation convention holds.
In the ‘usual’ applications to normal hydrodynamic situations, the velocity ui is a
single-valued, smooth function of xi and (for laminar flow, at any rate) gives no sign
of irregularity. However, there are situations in which irregularities do appear in the
velocity field and, as these irregularities grow, the turbulent situation is approached.
The following physical example will clarify this last point.
Consider two coaxial circular cylinders, with a fluid contained between them.
Let ωo and ωi be the angular velocities of the outer and inner cylinders, respectively.
Consider also that ωi > ωo . Then there exists a range of values E = ωi − ωo , ranging
from E = 0 to some critical value Ec , in which (1.1) and (1.2) are uniquely satisfied
by the time independent (stationary) solution
B
uθ = Ar + , ur = 0, uz = 0, (1.3)
r
E.A. Spiegel (ed.), The Theory of Turbulence, Lecture Notes in Physics 810, 1
DOI 10.1007/978-94-007-0117-5_1, © Springer Science+Business Media B.V. 2011
2 1 The Turbulence Problem
where (r, θ, z) are cylindrical coordinates and A and B are constants. But as E is
increased beyond Ec , the motion breaks up into a cellular pattern, and it appears
that other solutions [satisfying the boundary conditions as (1.3) does] can exist for
(1.1) and (1.2). At the first stage of this cellular motion only one cell size exists; this
is the one which dissipates energy least effectively. But as E increases, more and
more cell sizes appear until a spectrum of sizes exists.
These new cellular motions in the fluid represent various new solutions of (1.1)
and (1.2). It may be shown that, if the Reynolds number R = ρvl/μ is sufficiently
large, numerous solutions to (1.1) and (1.2) may be found to satisfy a particular
set of boundary conditions. As more and more solutions become possible in any
physical situation (and therefore simultaneously manifest themselves) a situation
of ‘fully developed turbulence’ is approached. We see that, from this standpoint,
‘turbulence’ refers to an idealized situation, and that fully developed turbulence can
only be approximated by any physical situation.
1.2 Two Fundamental Aspects of Turbulence
(a) Viscous Dissipation of Energy. If energy is supplied to a system at a fixed rate,

the system is in an essentially stationary state if this energy can be dissipated
at the rate (on the average) at which it is supplied. (Of course, it remains to
describe the sense in which the system is stationary.) Viscous dissipation is the
only mechanism available to a fluid medium to dissipate the energy input and
thus maintain an energy balance. Because of this, and because it is physically
clear that turbulence cannot exist in an inviscid medium, viscosity and viscous
dissipation are necessary aspects of the turbulence phenomenon.
(b) Interchange of Energy Between the Various States of Motion. We have seen
that the existence of turbulence in a physical situation implies the coexistence
of states of motion representing various solutions of (1.1) and (1.2) and subject
to the boundary conditions. If these states of motion did not somehow interact,
the turbulence problem would resolve (or degenerate) into the problem of de-
scribing these several states of motion independently. However, an interaction
does occur between the states of motion in the form of a continual interchange
of energy, as we shall see in Chap. 3.
The properties (a) and (b) are fundamental to the turbulence problem which may be
framed in terms of them, namely: what is the mechanism of (b), and how may we,
in view of the effects of (a) and (b), describe the phenomenon that we have called
turbulence?
Chapter 2
The Net Energy Balance
In this section we see what the terms in the Navier–Stokes equation contribute to
the production and dissipation of energy in a fluid. Consider a fluid contained in a
volume V whose boundary is the surface S. We prescribe that no material shall cross
S and we thus have the boundary condition that the velocity component normal to
S is zero everywhere on S, i.e.,
u · dS = 0 on S, (2.1)
where dS is a vector normal to the surface and of arbitrary magnitude. If we multiply

(1.1) by ρui we obtain, after integrating both sides over V ,

∂ui ∂ui ∂p
ρ ui dV + ρ ui uj dV = − ui dV + νρ ui ∇ 2 ui dV . (2.2)
V ∂t V ∂xj V ∂xi V
Here dV is the element of volume in V , ρ has been treated as constant, and the
solenoidal character of ui will be assumed throughout. We then consider the four
terms of (2.2) separately:
(a) The First Term on the Left-Hand Side:

∂ui ρ ∂(ui ui ) ρ ∂|u|2 ∂T
ρ ui dV = dV = dV = ,
V ∂t 2 V ∂t 2 V ∂t ∂t
where T is the kinetic energy contained in the volume.

(b) The Non-Linear Term: In (1.1) the term ∂(ui uj )/∂xi arises as the changing
velocity of a mass element arising from its changing position in the velocity
field; it is known as the inertial term. We have,

∂ui 1 ∂(ui uj ) 1 ∂(uj |u|2 )
ρ ui uj dV = ρ uj dV = ρ dV ,
V ∂xj 2 V ∂xj 2 V ∂xj
4 2 The Net Energy Balance
since uj is solenoidal. Then, by the divergence theorem,

∂ui 1
ρ ui uj dV = ρ div (u|u|2 )dV
V ∂xj 2 V

1
= ρ |u|2 u · dS
2 S
= 0 by (2.1).
(c) The Pressure Term:

∂p ∂(pui )
ui dV = dV = pu · dS = 0,
V ∂xi V ∂xi S
by (2.1).
(d) The Dissipation Term: We have the lemma
ui ∇ 2 ui = −|curl u|2 + div (u × curl u),
proved at the end of this section. Then, in view of this identity,

ρν ui ∇ ui dV = μ
2
− |curl u|2 + div (u × curl u) dV
V V

= −μ |curl u|2 dV + μ (u × curl u) · dS.
V S
Let ω = curl u. Then, on gathering the various terms of (2.2) we obtain

∂T
= −μ |ω|2 dV + μ (u × curl u) · dS. (2.3)
∂t V S
The first term on the right-hand side is the viscous dissipation of the vorticity, i.e.,
−μ|ω|2 is the rate of dissipation of energy per unit volume. Thestationary state
requires ∂T /∂t = 0, and so the energy input must be balanced by V (−μ|ω|2 )dV .
This implies that the small scale motion predominates in the dissipation, as will
become clear below.
Proof of the Lemma

To show
ui ∇ 2 ui = −|curl u|2 + div (u × curl u),
consider
∂uk
curli u = ij k , |curl u|2 = ij k imn uk,j un,m . (2.4)
∂xj
But
ij k imn = δj m δkn − δj n δkm . (2.5)
2 The Net Energy Balance 5
Combining (2.4) and (2.5) leads to
|curl u|2 = (δj m δkn − δj n δkm )uk,j un,m

= ui,j ui,j − ui,j uj,i .
Further,
(u × curl u)i = ij k uj kmn un,m

= (δim δj n − δin δj m )uj un,m = uj uj,i − uj ui,j , (2.6)
whence
div (u × curl u) = uj,i uj,i + uj uj,ii − uj,i ui,j − uj ui,j i .
Now,

∂ ∂ui
uj ui,j i = uj ui,ij = uj = 0,
∂xj ∂xi
by (1.2). Then,
div (u × curl u) − |curl u|2 = ui ∇ 2 ui ,
and the lemma is proved.
Returning to the example of concentric rotating cylinders, we note that the second
term on the right of (2.3) must be the energy introduced by the cylinders per unit
time and must be balanced by the dissipation term if ∂T /∂t is to be zero. Let i, j , k
be unit vectors in the r, θ, z directions, respectively. Then, the velocity (at least in
one solution) is given by vr = vz = 0 and vθ = v, where v is given by (1.3). Then

1 ∂(rv)
curl u = k = 2Ak,
r ∂r
whence
u × curl u = i2Av,
and
(u × curl u) · dS = 2AvdS,
so that
l 2π R2
(u × curl u) · dS = 2Av rdθ dz
0 0 R1

B B
= 4πlA R2 AR2 + − R1 AR1 +
R2 R1
= 4πlA2 (R22 − R12 ).
6 2 The Net Energy Balance
Also, since

1 ∂(rv) 2
|curl u|2 = = (2A)2 ,
r ∂r

|curl u| dV = 4A
2 2
dV = 4A2 (πlR22 − πlR12 )
V V
= 4πlA (R22 − R12 ),

2
and ∂T /∂t = 0, which will hold on the average, even when instability occurs.
Chapter 3
Interchange of Energy Between States of Motion
That energy is continually interchanged between the various states of motion was
stated in Sect. 1.1. We now state this property of turbulence analytically in the lan-
guage of Fourier analysis.
Maintaining our notion of an incompressible fluid contained in a finite volume,
we make a Fourier representation of the velocity field of the medium:

u(r, t) = uk (t)eik·r . (3.1)
k
Here k is the wave vector, corresponding to the wavelength

π
λ= (3.2)
|k|
and the Fourier transform uk is the velocity of the component represented by k in
the resolution.
On applying (1.2) to (3.1) we may observe that

div u = i (uk · k)eik·r = 0. (3.3)
k
Equation (3.3) will be true in general (only) if
uk · k = 0. (3.4)
The orthogonality of uk and k implies that the time variation of uk is a rotation in

the plane orthogonal to k.
For the sake of brevity, let
p
= , (3.5)
ρ
and perform the Fourier resolution of :

(r, t) = k (t)eik·r . (3.6)
k
8 3 Interchange of Energy Between States of Motion
Another convenient step is to rewrite (1.1) in vector form. We then have

∂u
+ (u · grad)u = −grad + ν∇ 2 u. (3.7)
∂t
Introducing (3.1) and (3.6) into (3.7), we find
∂ ∂
uk (t)eik·r + (uj )k eik·r u eik ·r
∂t ∂xj k
k k k

= −i k keik·r − ν uk |k|2 eik·r . (3.8)
k k
The second term on the left-hand side of (3.8) may be simplified. In this term, the
summation indicated by the repeated index j should not be overlooked. That term
becomes

uk · (k − k )uk−k eik·r ,
k k
with the use of the standard formula for the products of infinite series, namely
∞
∞
∞
∞
an z n bn z n = ak bn−k zn .
n=0 n=0 n=0 k=0
Finally, under application of (3.4), the term becomes

i uk−k (uk · k)eik·r .
k k
The resolved form of (3.7) in final form is thus

∂uk
eik·r + i uk−k (uk · k)eik·r
∂t
k k k

= −i k keik·r − ν uk |k|2 eik·r . (3.9)
k k
If the eik·r coefficients are equated term by term, there results

∂uk
+i uk−k (uk · k) = −i k k − ν|k|2 uk . (3.10)
∂t
k
We may multiply both sides of (3.10) by u∗k to get the rate of change of kinetic
energy for a given k. Then
1 ∂|uk |2
= −i uk−k (uk · k) · u∗k − i k k · u∗k − ν|k|2 |uk |2 . (3.11)
2 ∂t k
3 Interchange of Energy Between States of Motion 9
The rate of change of energy for a given k is thus seen to be the sum of three terms:
(1) The term

Q(k , k) = −i uk−k (uk · k) · u∗k
k k
gives the rate of interchange of energy between the Fourier component k and all
other components k . The quantity Q(k , k) is a sort of transition probability,
giving relative strengths of the tendency for energy to be exchanged with k by
the various k . This interaction term comes from the inertial term in (1.1), and
so (see Sect. 1.2) cannot alter the net energy balance; its average effect over all
k must be null. (It should also be remarked that stability problems say nothing
of the interaction between components since they depend on the neglect of the
inertial term.)
(2) The term −ik k · u∗k has no interaction properties, but can cause the exchange
of energy between the various spatial positions of the same Fourier component,
i.e., it will spread the energy spatially, but will maintain it in fixed k-space
position. It is intuitively likely that this term will tend to promote isotropy. For
the moment, we will assume isotropy, and neglect this term.
(3) The term −ν|k|2 |uk |2 represents the viscous dissipation by the k component. If
(1.1) and (1.2) were not present in (3.11), this term would lead to a solution of
the form
uk (t) = uk (0)e−ν|k| t .
2
To sum up, only the inertial term contributes to the energy interchange. The balance
of energy for any component k is governed, in the case of isotropy, by the equation
1 d|uk |2
= Q(k , k) − ν|k|2 |uk |2 . (3.12)
2 dt
k
Chapter 4
Some Remarks
4.1 On the Harmonic Analysis
We have heretofore considered finite volumes over which to Fourier-analyze veloc-

ities and pressures. Since finite volumes will have only a finite number of modes,
Fourier series have been adequate for these pruposes. But for the analysis of an
infinite region, the Fourier integral is needed. We introduce the latter at this point.
The number of modes per volume of k-space of size dVk = dkx dky dkz is
V
dn = dkx dky dkz . (4.1)
(2π)3
This may be best visualized in one dimension: a length contains
lx
nx = (4.2)
λx
modes. On use of (4.1), the number of modes in the range kx to kx +dkx is, therefore,

lx kx lx
dnx = d = dkx . (4.3)
2π 2π

On the basis of (4.1) the operator k becomes

V
→ dk dky dkz .
3 x
k k (2π)
4.2 On the Concept of Isotropy
It is initially clear that in isotropy we should be interested in absolute values con-

nected with k, and not in associated directions. We will, for the moment, content
ourselves to make use of this notion of isotropy without exploring it further. The
12 4 Some Remarks
notion implies that for isotropy we must have spherical symmetry in k-space. Then,
the number of modes in a volume V in physical space contained in a range dk of k
is (see Richtmeyer and Kennard, p. 188)
V
dn = 4πk 2 dk. (4.4)
(2π)3
The Fourier integral operator then becomes
∞
V
4πk 2 dk.
0 (2π)3
4.3 On the Possibility of a Universal Theory
We saw in Chap. 2 that in, the simple case of concentric, rotating cylinders, energy
was introduced at the boundary and dissipated throughout the volume. Such a situ-
ation implies an inherent inhomogeneity which is characteristic of turbulence as it
occurs in nature.
The question then arises: Does there exist a volume element, i.e., some physical
domain, sufficiently small (but not so small as to be insignificant) and sufficiently far
from the boundary that it may be considered independent of the nature of conditions
at the boundaries and of the inhomogeneity they introduce? If such a domain can be
defined, the laws which apply to the turbulence in it will be of universal character.
Chapter 5
The Spectrum of Turbulent Energy
5.1 The Spectrum
We may obtain the kinetic energy of motion (the square of the velocity) by evaluat-
ing u · u∗ . We then obtain

|u|2 = uk eik ·r · u∗k e−ik ·r
k k

= uk · u∗k e−i(k −k )·r
. (5.1)

k k
The mean-squared velocity in the volume V is

1
|u| =
2 e−i(k −k )·r uk · u∗k dV
V V
k k

1
= uk · u∗k ei(k−k )·r dV ,
V V
k k
where we have replaced k by k.

The integral in the last term vanishes for k − k = 0, so

1
|u|2 = uk · u∗k dV ,
V V k
or

|u|2 = |uk |2 . (5.2)
k
The summation of (5.2) can be converted to an integration by the considerations of

Sect. 4.2, if V → ∞. The mean-squared velocity, averaged over all space, is then,
14 5 The Spectrum of Turbulent Energy
for the case of isotropic turbulence,

∞
V
|u|2 = u2 = 4πk 2 |uk |2 dk,
(2π)3 0
and the mean kinetic energy per unit mass is

∞ ∞
1 2 V V
u = k |uk | dk =
2 2
k 2 |uk |2 dk. (5.3)
2 (2π)2 0 (2π)2 0
The function
V
F (k) = k 2 |uk |2 (5.4)
(2π)2
is the spectrum of (isotropic) turbulence.
For the non-isotropic case, a three-dimensional spectrum may be employed.
Then,

V
|u|2 = |uk |2 dkx dky dkz , (5.5)
(2π)3 k
and the general turbulence spectrum is
V
F (k) = |uk |2 . (5.6)
(2π)3
5.2 An Equation for the Spectrum
Any equation that the energy spectrum satisfies is likely to follow from (3.7), (1.1),
and (3.6), which we recall here:
∂u 1
+ (u · grad)u = − grad p + ν∇ 2 u, (5.7)
∂t ρ

u(r, t) = uk (t)eik ·r , (5.8)
k
and

= k eik·r . (5.9)
k
Since u(r, t) is real,

u(r, t) = u∗ (r, t) ,
we have from (5.8) that
uk = u∗−k . (5.10)
5.2 An Equation for the Spectrum 15
(It might be noted here that the pressure term could simply be eliminated from
(3.10). Scalar multiplication by k, gives an expression for k which may be intro-
duced into the equation for k , but this step is unnecessary in what follows.)
From these equations, we found [see (3.10) on p. 8]
∂uk
= −i (uk · k)uk−k − i k k − ν|k|2 uk . (5.11)
∂t
k
Multiplying both sides of (5.11) by u∗k , we obtain, using (3.4),
∂uk
u∗k = −i (uk · k)(u∗k · uk−k ) − ν|k|2 |uk |2 . (5.12)
∂t
k
The complex conjugate of (5.12) is
∂u∗k
uk =i (u∗k · k)(uk · u∗k−k ) − ν|k|2 |uk |2 . (5.13)
∂t
k
Upon adding (5.12) and (5.13), we find that
∂|uk |2
= −i (uk · k)(u∗k · uk−k ) + i (u∗k · k)(uk · u∗k−k ) − 2ν|k|2 |uk |2 .
∂t
k k
(5.14)
In the first summation of (5.14), we replace k by k − k . It then becomes

(uk−k · k)(u∗k · uk ). (5.15)

k−k

For a fixed k, k−k is equivalent to summation over −k , viz., −k , so (5.15)
is equivalent to

(uk−k · k)(u∗k · uk ). (5.16)
−k
We may replace the index of summation −k by a more convenient one, say k .

Then (5.16) becomes, when (5.10) is introduced,

(uk+k · k)(uk · uk )∗ . (5.17)
k
In a similar way, we may replace k by k + k in

(u∗k · k)(uk · u∗k−k ) (5.18)
k
16 5 The Spectrum of Turbulent Energy
to obtain

(u∗k+k · k)(uk · u∗−k ), (5.19)

k
into which the same reasoning has been introduced as in going from (5.15) to (5.16).
Replacing k by k in (5.19) and introducing (5.10), we find

(u∗k+k · k)(uk · uk ). (5.20)
k
Then (5.20) and (5.17) are the equivalent of the sums in (5.14), and if we let

Q(k, k ) = i (u∗k+k · k)(uk · uk ) − (uk+k · k)(uk · uk )∗ , (5.21)
we have, for fixed k [see (3.12)],

1 d 1
|uk |2 = − Q(k, k ) − ν|k|2 |uk |2 . (5.22)
2 dt 2
k
The foregoing equation has been discussed above. It gives the rate of change of
kinetic energy in a Fourier component. This rate of change has two components:
(a) A Damping Term. This term alone gives rise to an exponential decay in a given
Fourier component with a decay time
1
τ= . (5.23)
2ν|k|2
Equation (5.22), as a result of the linearity of the velocity derivative in the
Navier–Stokes equation, shows that the damping in one k is independent of
all other k. Moreover, shorter wavelengths decay most quickly, the decay time
being proportional to λ2 .
The λ2 dependence leads to an interesting property of decaying turbulence.
If there is turbulence which is stationary and isotropic in some volume V , and
the source of energy is removed, the isotropy will be destroyed with time. This
occurs because the smaller eddies, which lend isotropy to the turbulence, decay
first, leaving the larger, boundary-dependent eddies to exist for longer times.
The larger eddies are, of course, anisotropic.
(b) The Transfer Term. The term Q(k, k ) is a sort of transition probability for the
transfer of energy between the k and k components. But Q(k, k ) depends on
the component k + k , the intermediary through which the energy must pass.
As would be expected from conservation reasoning, Q(k, k ) is antisymmet-
ric, i.e.,
Q(k, k ) = −Q(k , k). (5.24)
This last remark follows from the definition of Q(k, k ) and the property (3.4),
viz.,
uk+k · (k + k ) = 0. (5.25)
5.2 An Equation for the Spectrum 17
For if k and k are interchanged in Q as given by (5.21), uk+k · k replaces

uk+k · k. But by (5.25),
uk+k · k = −uk+k · k, (5.26)
and (5.24) follows.

We may now proceed to see how (5.22) provides an equation for the turbulence
spectrum. Assuming isotropy, we multiply (5.22) by k 2 and, on noting (5.4), we
find
∂F (k, t) V 2
=− k Q(k, k ) − 2ν|k|2 F (k, t). (5.27)
∂t (2π)2
k
If we pass from the sum to the integral in (5.27), we get
∞
V
k 2 Q(k, k ) → 3
4π k 2 k 2 Q(k, k )dk,

(2π) 0
k
and (5.27) becomes

∞
∂F (k, t)
− = T (k, k )dk + 2ν|k|2 F (k, t), (5.28)
∂t 0
in which
V 2 2 2
T (k, k ) = 2 × k k Q(k, k ). (5.29)
(2π)4
From (5.28), one would hope to develop a theory of turbulence. Two procedures
seem possible:
1. Develop some physical description of turbulence and justify the results in terms
of (5.28).
2. Carry (5.28) as far as possible mathematically, and give a physical interpretation
to the results.
Procedure (1) has been followed in the past, and we will here consider Heisenberg’s
scheme as it embodies most others which proceed along those lines. Following this,
we will show how procedure (2) may be exploited.
Chapter 6
Some Preliminaries to the Development
of a Theory of Turbulence
We may integrate (5.28) over the wave numbers 0 to k :

k k ∞ k
1 ∂
− F (k, t)dk = dk T (k, k )dk + ν k 2 F (k)dk, (6.1)
2 ∂t 0 0 0 0
where a factor of 1/2 has been absorbed into the definition of T (k, k ). The expres-
sion (6.1) gives the rate of passage of energy from the eddies included in the range
from 0 to k.
The integral over k on the right-hand side of (6.1) may be simplified by intro-
ducing the antisymmetric property of Q(k, k ). It is in fact clear that T (k, k ) is also
antisymmetric, i.e.,
T (k, k ) = −T (k , k). (6.2)
Hence,
k k
dk T (k, k )dk = 0, (6.3)
0 0
and (6.1) becomes
k k ∞ k
1 ∂
− F (k, t)dk = dk T (k, k )dk + ν k 2 F (k, t)dk. (6.4)
2 ∂t 0 0 k 0
On the right-hand side:
1. The first integral gives the amount of energy flowing per unit time into the eddies
in the range k to ∞ from those in the range 0 to k.
2. The second integral gives the rate at which energy is dissipated by viscosity in
the range 0 to k. To aid in the understanding of the structure of the dissipation
term, we will investigate the way it may arise from the energy balance equation.
We saw that the mean rate of dissipation over the volume V is [see (2.3) on p. 4]

1
ε = −μ |curl u|2 dV = −μ|curl u|2 . (6.5)
V V
20 6 Some Preliminaries to the Development of a Theory of Turbulence
If we put the Fourier expression for u into (6.5), we will discover the contributions
to the dissipation by the various Fourier components.
On p. 4, we wrote the relation
∂un
curli u = imn ,
∂xm
where u may be replaced in accord with
∂
curli ū = imn (uk )n eik·r
∂xm
k

= imn i (uk )n km eik·r .
k
But
imn km (uk )n = (k × uk )i ,
so

curl u = i (k × uk )eik·r . (6.6)
k
Moreover,

(curl u)∗ = −i (k × u∗k )e−ik·r . (6.7)
k
Then

|curl u|2 = (k × uk )∗ · (k × uk )ei(k−k )·r . (6.8)
k k
If we integrate (6.8) over a volume which we will let grow to include all r, the

ei(k−k )·r will go to zero unless k = k , and the Fourier summation will tend to an
integral. Then
∞
V
ε= 4πk 2 |k × uk |2 dk. (6.9)
(2π)3 0
But by the orthogonality of k and uk , we find
|k × uk |2 = |k|2 |uk |2 ,
and
∞
ε = 2ν F (k)|k|2 dk, (6.10)
0
whence the dissipation by just those eddies in the range 0 to k is clearly
k
εk (thermal) = 2ν |k|2 F (k)dk. (6.11)
0
Chapter 7
Heisenberg’s Theory of Turbulence
7.1 The Fundamental Equation of the Theory
We have seen that the rate at which energy passes from the range 0 to k of wave
numbers into the range k to ∞ is
k
1 ∂
εk = − F (k, t)dk, (7.1)
2 ∂t 0
whence according to (6.4)
k ∞ k
εk = dk T (k, k )dk + ν k 2 F (k)dk. (7.2)
0 k 0
From the physical interpretation of
k ∞
εk (mechanical) = dk T (k, k )dk , (7.3)
0 k
namely that the eddies in the range k to ∞ take mechanical energy from those
in the range 0 to k, arises the notion that εk (mechanical) must have the form of a
dissipation term in which the dissipation in the range 0 to k is provided by the eddies
in k to ∞. Heisenberg therefore assumes that εk (mechanical) is of the form
k
εk (mechanical) = νk F (k )k 2 dk , (7.4)
0
in analogy with
k
εk (thermal) = ν F (k )k 2 dk , (7.5)
0
where νk is the ‘viscosity’ of the eddies k to ∞ acting on 0 to k.
Another assumption made in this theory is that T (k, k ) is separable in k and k ,
i.e.,
T (k, k ) = f (k)g(k ). (7.6)
22 7 Heisenberg’s Theory of Turbulence
By our assumption (7.4), we see that f (k) must be given by

f (k) = k 2 F (k). (7.7)
It is also plausible that g(k ) should be expressible in terms of k and F (k ). More-
over, g(k ) will give the νk and it is clear, both by analogy with (7.3) and from the
notion that νk is a ‘viscosity’ produced by the k in the range k to ∞, that
∞ k
εk (mechanical) = g(k )dk k 2 F (k)dk. (7.8)
k 0
And finally, consistently with the physical ideas we have introduced, we must ensure
∞
that k g(k )dk has the same dimensions as ν. If we let [ ] operating on some
quantity mean ‘dimensions of’, we have
L2 L3
[F ] = 2
L= 2, (7.9)
T T
and
L2
[ν] = . (7.10)
T
We must therefore have
L2
[g][dk] =
T
and
L3
[g] =
. (7.11)
T
Since we have assumed that g is expressible in terms of (powers of) F (k ) and k ,
we must also have
L3
[F ]α [k]β = , (7.12)
T
or
L3α−β T−2α = L3 T−1 , (7.13)
whence

3α − β = 3,
(7.14)
−2α = −1.
Then (7.14) has the solutions
1 3
α= , β =− , (7.15)
2 2
and we find for g(k ) the result

F (k )
g(k ) = . (7.16)
k 3
7.2 Chandrasekhar’s Solution of (7.17) for the Case of Stationary Turbulence 23
At this point we change the definition of the ε to conform to the literature without
altering the notation. In order to make ε the rate of flow of energy per unit mass,
rather than per unit volume, we introduce the factor ρ (density) into all the ε. More-
over, we will bring the factor 1/2 in (7.1) to the other side, thus multiplying the
various expressions for ε by 2. With these considerations and equations (7.5), (7.8)
and (7.16), the rate of flow of energy per unit mass of fluid from the eddies in the
range 0 to k is
∞
F (k ) k
εk = 2ρ ν + K0 dk F (k)k 2 dk. (7.17)
k k 3 0
In the case of stationary turbulence with sufficiently high Reynolds number,

εk = constant, (7.18)
while for the decay of turbulence,
k
∂
εk = − F (k, t)dk. (7.19)
∂t 0
From (7.17) may be derived the results of Kolmogorov and the rest of this school of
similitude. Perhaps the chief contribution of Heisenberg is that he has embodied all
the previous attempts along this line in equation (7.17).
7.2 Chandrasekhar’s Solution of (7.17) for the Case of

Stationary Turbulence
In the case of stationary turbulence in which viscosity is assumed to act only on the
largest k values, (7.18) applies. Then differentiation of (7.17) with respect to k gives
∞
F (k ) F (k) k 2
νF (k)k + K0 k F (k)
2 2
dk − K0 k F (k)dk = 0. (7.20)
k k 3 k3 0
We may rearrange (7.20) to give
∞ k
ν dk ) =
1
+ F (k
1/2 F (k)k 2 dk. (7.21)
K0 k k 3/2 2
k F (k)k 3 0
We let
⎧
⎨ g = k F (k),
3
⎪
k (7.22)
⎪
⎩y = F (k)k 2 dk.
0
Then,
dy g
= , (7.23)
dk k
or
dy dk
= ,
g k
so that, if we can get g(y), we can find k :
y dy
log k = constant + . (7.24)
0 g(y)
Introducing (7.22) and (7.23) into (7.21), we find
∞
ν dy y
+ 1/2 2
= 2√ . (7.25)
K0 y g k k g
If we differentiate (7.25) and use (7.24), taking y as the independent variable,

1 1 d y 2y dk
− 1/2 2 − 2 √ + 3√ = 0,
g k k dy g k g dy
or

1 1 d y 2y
− 1/2 2 − 2 √ + 2 3/2 = 0. (7.26)
g k k dy g k g
Carrying out the differentiation, we find
1 1 1 y dg 2y
− √ 2 − √ 2 + 2 3/2 + = 0,
gk gk 2k g dy k 2 g 3/2
or
dg g
− 4 + 4 = 0. (7.27)
dy y
The solution of the homogeneous form of (7.27), i.e., of
dg g
− 4 = 0, (7.28)
dy y
is
g = Ay 4 , (7.29)
where A is an integration constant, and a particular integral of (7.27) is
g = By. (7.30)
Putting (7.30) into (7.27), we find that B must satisfy
B − 4B + 4 = 0, (7.31)
whence
B = 4/3. (7.32)
The general solution is therefore
4
g = Ay 4 + y,
3
or
4
g = y(1 − Cy 3 ). (7.33)
3
We may now return to the integral in (7.24). It is
y y
dy dy 1 y d(y 3 )
= = .
0 3 y(1 − Cy 3 ) 3 0 43 y 3 (1 − Cy 3 )
4
0 g
The foregoing is

1 y d(y 3 ) u du
= .
4 0 y (1 − Cy 3 )
3
0 u(1 − Cu)
By the method of partial fractions, we may write
M N 1
+ = ,
u 1 − Cu u(1 − Cu)
whence
M(1 − Cu) + Nu = 1.
Setting u = 0 and u = 1/C, we find that
M = 1, N = C.
Then
u u u
du du d(Cu)
= +
0 u(1 − Cu) 0 u 0 1 − Cu
∝ log y 3 − log(1 − Cy 3 )
y3
= log . (7.34)
1 − Cy 3
Combining (7.24) and (7.34), we find
1/4
y3
k=α , (7.35)
1 − Cy 3
where α is an integration constant. On solving (7.35) for y, we obtain
k4
(1 − Cy 3 ) = y 3 , (7.36)
α4
k4
y3 = . (7.37)
α4 + Ck 4
Or, on substituting 1 − Cy 3 as found from (7.36) into (7.33), we have
4 αy 4
g= . (7.38)
3 k
Then the spectrum is [see (7.22)]
g
F (k) =
k3
4 α4 4
= y [by (7.38)]
3 k7
4/3
4 α4 k4
= [by (7.37)] .
3 k 7 α 4 + Ck 4
Finally,
4 α4
F (k) = . (7.39)
3 k 5/3 (α 4 + Ck 4 )4/3
This expression may also be conveniently rewritten by factoring out α 4 :
4 α −4/3 1
F (k) = . (7.40)
3 k 5/3
1 + C4 k 4 α
If we write
−1
C 1/4 4 1
ks = , F (k0 ) = , (7.41)
α 3 α 4/3 k 5/3
0
equation (7.40) becomes
5/3
k0 1
F (k) = F (k0 ) . (7.42)
k [1 + (k/ks )4 ]4/3
In obtaining the solutions of the Heisenberg equation, differentiations were per-
formed which raised the order of the equations. Thus, an additional integration con-
stant was introduced. It is therefore to be expected that α and C are not independent,
and in particular, that they must be related through the differential equation (7.25).
If (7.39) and (7.22) are introduced into (7.25), or its equivalent (7.21), we find
k
ν y α2 1 d 1
= 2 1/2 − √ dk. (7.43)
K0 k g 2 3 ∞ (C + α /k ) dk k 4
4 4 2/3
The integral in (7.43) is, under the substitution u = 1/k 4 ,

u u
du 3 4 1/3
= + (C + α u)
0 (C + α u)
4 2/3 α4 0
1/3
3 α4 1/3
=+ 4 C+ 4 −C ,
α k
and (7.43) is therefore
√ 1/3
ν y 3 α4
= − C+ 4 − C 1/3 . (7.44)
K0 k 2 g 1/2 2α 2 k
By (7.38) we have
√
y y 3
= = 2 , (7.45)
k 2 g 1/2 k 2 √2 α 2 y 2 2α y
3k2
and by (7.37),
1/3
1 (α + Ck 4 )1/3 α4
= = C+ 4 . (7.46)
y k 4/3 k
Then (7.44) becomes
√
ν 3 1/3
= C . (7.47)
K0 2α 2
√ 3/4
1 3 K0
ks = √ . (7.48)
α 2 ν
We note that, as ν → 0 (and therefore k → ∞), ks → ∞. We see that in (7.42),
5/3
k0
F −→ F (k0 ) as ν → 0, (7.49)
k
and we have recovered the Kolmogorov k −5/3 law. On the other hand, when ν is not
zero, and k ks ,
F ∝ k −7 . (7.50)
Actually, (7.50) is not verified by experiment.
In conclusion, we have found that the following picture presents itself:
1. There is a k0 beyond which F (k) decreases as k increases, so that beyond k0 ,
i.e., k > k0 , the spectrum will not depend on the boundary conditions, it being
physically clear that L0 = 2π/k0 is the dimension of the vessel. In the region
beyond k0 , the spectrum follows a k −5/3 power law.
2. There exists a k = ks at which the viscous term is comparable with the inertial
term. As k increases beyond ks , the viscous term predominates and a k −7 law
obtains.
The region from k0 to ks is called the universal equilibrium range.
Chapter 8
Other Derivations of the k −5/3 Law
The methods in this section are based on the assumption that local isotropy exists.
Local isotropy may be thought of as the existence of a k0 such that the boundary
conditions are irrelevant for k > k0 . These methods also require that for k → ∞
there is a finite range k0 k ks , where ks has the meaning of the previous section.
8.1 Fermi’s Approach
For an eddy of wave number k [see (6.6) on p. 20],

εk = νk |curl v k |2 = νk (vk k)2 .
Since the dimensions of ν are L2 /T, we must have
vk
νk ∼ ,
k
so
εk ∼ vk3 k = const. for k0 k ks .
Now this last expression indicates
vk ∼ k −1/3 .
But we know that [F (k)] = L3 /T2 so
vk2 1
F (k) ∼ = 5/3 . (8.1)
k k
Since we have been dividing by k, especially in the preceding equation, the results
must apply to all eddies in the range ∞ to k, since (eddy)k contains the rest. (That
is, it contains the ones in k to ∞.)
Now the inflow from k < ks into k > ks is
νk |curl v|2 ,
30 8 Other Derivations of the k −5/3 Law
where curl v refers to the velocity of the eddies 0 to k. By the reasoning of the
preceding sentences,
vk
νk ∼ ,
k
where νk refers to all the eddies in 0 to k. Moreover, since curl v refers to all these
eddies, and since k ∼ ks , it is (nearly) the curl of the actual velocity. It must therefore
be constant, on the average. So the inflow of energy to the range k > ks is (especially
as ks → ∞)
vk
constant × .
k
Moreover, the outflow due to dissipation is
vk
ν|curl vk |2 ∼ νvk2 k 2 = inflow = × constant.
k
Furthermore, vk ∼ k −3 , whence
F (k) ∼ k −7 . (8.2)
8.2 Kolmogorov’s Theory

The theory of Kolmogorov follows from two principles of similitude:
1. There exists a range of eddy sizes k > k0 in which the spectrum, and other distri-
bution functions of physical character, do not depend on the boundary conditions
in any way. In this range, for finite Reynolds numbers, these distribution func-
tions depend only on ν and εk .
2. In the limit as ν → 0, all these functions will depend only on εk . The action of ν
becomes relegated to increasingly larger k values.
It would follow that there exist certain universal functions characterizing the turbu-
lence in those parts of Fourier space unaffected by the boundary conditions of the
individual problem. This sort of similarity principle can exist if the lengths involved
are expressed in an appropriate dimension. Now, a length η can be constructed from
the fundamental parameters ν and k by taking
3 1/4
ν
η= , (8.3)
ε
since [ν] = L2 /T and [ε] = L2 /T3 . Moreover, velocities can be constructed from
the unit
[velocity] = (νε)1/4 . (8.4)
In these units, i.e., (8.3) and (8.4), all functions may have universal character. For
example, the spectrum function can be constructed:
1/4
F (k) = ν 5/4 ε1/4 f k ν 3 /ε , (8.5)
8.3 The Method of von Neumann 31
where f is a universal function (assuming that there is local isotropy).

According to Kolmogorov’s second principle, when ν → 0, F (k) must be inde-
pendent of ν, so that as ν → 0, f must be proportional to ν −5/4 . That is,
3 −5/12
−5/3 ν
f → Ck as ν → 0, (8.6)
ε
where C is a proportionality constant, so that for large Reynolds numbers,
F = Cε2/3 k −5/3 , (8.7)
which is the well-known Kolmogorov spectrum.
Of course, these dimensional arguments are based on the assumption that the
various functions describing the spectrum follow power laws. If this is not the case,
these arguments break down.
8.3 The Method of von Neumann
As we have seen, the two integration constants which arose in the Chandrasekhar
solution of the Heisenberg equation are related through the viscosity. Thus k0 and
ks as they arise in Heisenberg’s theory should be related through ν. It should also
be asked how ks behaves as the Reynolds number tends to infinity, and how the
ratio ks /k0 depends on the Reynolds number. From the Heisenberg theory, we find
(approximately)
3/4
ks = 0.2211K0 R0 K0 , (8.8)
where R0 is the Reynolds number of the entire motion [see PRS 200, 20 (1949),
3/4
equation (27)]. That ks /k0 should go as R0 , as well as the need for definitions of
quantities such as k0 and ks appears in von Neumann’s theory, as we shall now see.
The mean energy per unit mass is
1
E = u2 ,
2
which is strictly
∞
E= F (k)dk. (8.9)
0
The energy dissipation per unit volume is strictly
∞
W = 2ν F (k)k 2 dk. (8.10)
0
If we wish to accept the results of Kolmogorov we must replace F (k) by k−5/3 .

(Proportionality constants are overlooked here.) Then these expressions become
∞
E= k −5/3 dk (8.11)
0
32 8 Other Derivations of the k −5/3 Law
and
∞
W = 2ν k 1/3 dk. (8.12)
0
On physical grounds we should expect E < ∞ and W < ∞, but this is not the case
according to these last expressions. The integral (8.11) diverges for small k and
the integral (8.12) diverges for large k. In order to provide for the convergence of
(8.11), we must assume a lower limit for the integration which allows convergence,
but which does not alter the accuracy of the expression too greatly. That is, we want
a k0 = 0, but such that the eddies k < k0 contain little of the energy. (Physically, one
feels that k0 = 2π/L0 , where L0 is a dimension of the vessel.) If such a k0 exists,
then to a tolerable approximation, we have
∞
E= k −5/3 dk. (8.13)
k0
Similar reasoning applies to the integral (8.12). A suitable finite upper integration
limit ks must be found to ensure convergence. Then
ks
W = 2ν k 1/3 dk. (8.14)
0
Now the average energy of an eddy k is the average over the eddies whose wavenum-
bers we may take in the interval in [k0 , ks ]. This would be given by
ks
3
u2k = A k −5/3 dk = Ak −2/3 ,
k 2
where A is a suitable constant. Then
1/2
3
u2k = vk = A1/2 k −1/3 . (8.15)
2
The Reynolds number appropriate to k is
2π
k vk 2π( 32 )1/2 A1/2 k −4/3
Rk = = . (8.16)
ν ν
Now ks will occur when Rk is of order unity. (Not order 1000 because R = 1000
is needed to initiate turbulence, not to maintain it; experimentally, it is known that
once turbulence occurs, it persists to low Reynolds numbers ∼ 1.) Arbitrarily, we
shall choose
Rks = 2π. (8.17)
Then from (8.16), we have
4/3 (3/2)1/2 A1/2
ks = . (8.18)
ν
From (8.14) we see that
3 4/3
W = νAks , (8.19)
2
8.4 Conclusion 33
or, on dividing by (8.18) and rearranging, we have

2
A = W 2/3 . (8.20)
3
From (8.18), we find that
1/4
W
k= . (8.21)
ν3
On using (8.17) and (8.16), we have
−4/3
Rk0 Rk k
= 0 = 0−4/3 . (8.22)
Rks 2π ks
Hence,
3/4
Rk0
k s = k0 , (8.23)
2π
in agreement with the conclusion from the Heisenberg theory. This 3/4 power law
is also verified experimentally. Starting from this as an experimental result, it is
reasonable to conclude that, in terms of the Kolmogorov spectrum, viscosity takes
over at an Rk ∼ 2π .
8.4 Conclusion
The implication of the remarks in this and the last chapter is that you need only
specify the spectrum and the turbulence problem is solved. This remark is true for
the radiant energy distribution in the thermal enclosure, where phase relations do
not enter. However, in the turbulence problem the phase relations are fundamen-
tal; energy passes between the Fourier components, so that it will be necessary to
specify more than just the spectrum if the problem is to be solved.
Chapter 9
An Alternate Approach: Correlations
At the end of the previous chapter it was indicated that description of the spectrum
is not necessarily an adequate representation of turbulence. Some kind of statisti-
cal representation is what is usually sought; but this sort of search has underlying
difficulties.
Consider the statistical mechanics in kinetic theory. The Hamiltonian description
of a volume containing N molecules requires a 3N -dimensional description, to be
precise. The problem is, however, solved by a statistical representation in the 3-
dimensional physical space.
However, the Navier–Stokes equations give a precise representation in the phys-
ical 3-space, though the attempt is made to make a statistical representation in this
same space. That is, one is trying to represent statistically what the Navier–Stokes
equations represent exactly in the same framework.
One can ask whether it would be worthwhile to develop a new statistical me-
chanics. Perhaps it would, but in any case it is necessary to describe a fluctuating
velocity field, and we have seen that the spectrum is not necessarily fundamental
for such an attempt. Such a description might take the form of specifying the way
in which the velocity varies from point to point. This would be possible through the
correlations of velocities at different points. To do this we consider the velocities at
the points r and r + ξ :

1
ui (r) = ui (k)eik·r dk, (9.1)
(2π)3 k

1
uj (r + ξ ) = 3
uj (k )eik ·(r+ξ ) dk . (9.2)
(2π) k
Then the correlation is

ui (r)uj (r + ξ ) = uj (k )eik ·ξ dk ui (k)ei(k+k )·r dk. (9.3)
k k
The average value of ui (r)uj (r + ξ ) over space is then proportional to the integral
over space. Then (9.3) becomes

ik ·ξ
ui (r)uj (r + ξ ) = const. × uj (k )e dk ui (k)dk ei(k+k )·r dr.
k k V
36 9 An Alternate Approach: Correlations
The volume integral is a δ-function and vanishes unless k + k = 0. Using this con-
dition and a minor change in independent variable to k, we find

ui (r)uj (r + ξ ) = const. × ui (k)uj (k)eik·ξ dk. (9.4)
k
Or, if we have isotropy,
∞
ui (r)uj (r + ξ ) = const. × k 2 ui (k)uj (k)eik·ξ dk. (9.5)
0
Finally, if we set ξ = 0 and contract, we have
∞
|u(r)| = const. ×
2 |uk |2 dk, (9.6)
0
and we see that the correlations must be connected with the energy spectrum through
the Fourier transform.
Thus, in introducing the correlation function, we have introduced no essentially
new functions. However, it will be convenient to think in terms of correlations.
Moreover, the velocity correlations are more accessible to measurement than spec-
tral functions.
If we write uj for uj (r + ξ ), we have ui uj for the correlation. We will let
Qij (ξ ) = ui uj . (9.7)

Qij is a tensor since it is the product of two vectors. Also, if we let
ij (k) = ui (k)uj (k). (9.8)
equation (9.4) becomes

Qij (ξ ) = const. × ij (k)eik·ξ dk, (9.9)
where ij (k) is a generalized spectrum.

Chapter 10
The Equations of Isotropic Turbulence
For the case of isotropic turbulence, the spectrum specifies the turbulence. Since
Qij and ij are Fourier transforms of one another, it is to be expected that any
restrictions on one will imply restrictions on the other; but it must not be concluded
that one can pass back and forth between the two with ease. It is true that in working
with linear equations one can go to the Fourier space by means of an algebraic
equation, and in fact solutions are sometimes more readily obtained in this way.
But in nonlinear equations, one has to go from the products of transforms to the
transform of products, and this passage is usually quite complicated, so that the
nonlinearity of the Navier–Stokes equations inhibits the usefulness of the concepts
involved in passage to the Fourier space. Because of this difficulty, perhaps, the
preoccupation of many workers in turbulence with the Fourier space has been the
stumbling block in the way of a general theory.
10.1 The Concept of Isotropy
We say that turbulence is isotropic if the mean (over space, in this section) values of
quantities which are of scalar character and which involve the physical parameters
of the medium (such as velocity, pressure and density) are invariant under the full
rotation group, i.e., invariant under rotations as a rigid body and under reflections
through the origin.
The turbulence is homogeneous if these quantities are invariant under transla-
tions.
We will be interested in two kinds of correlation. We have already defined the
double correlation
Qij (ξ ) = ui uj . (10.1)
The triple correlation we will be interested in is
Tij k (ξ ) = ui uj uk . (10.2)
38 10 The Equations of Isotropic Turbulence
Fig. 10.1 Notation for

double correlation
Let the origin (or point of interest) be at P (see Fig. 10.1); at any rate let it be the
origin in the sense that we require invariance under reflection through P for isotropy.
And let P be displaced from P by ξ . Let a, b, c, . . . , be arbitrary unit vectors at P
and a , b , c , . . . , be arbitrary unit vectors at P . Then for isotropic turbulence it is
necessary that
ua ua = ai aj ui uj = ai aj Qij (10.3)
and
ua ub ua = ai bj ak ui uj uk = ai bj ak Tij k (10.4)
should be invariant under the full rotation group. With this restriction on Qij and
Tij k we can discover the forms of these tensors for isotropic turbulence.
Theorem Any invariant function of any number of vectors ξ , a, b, . . . , etc., can be

expressed in terms of the fundamental invariants of the following types:
• the scalar products such as ξ · a = ξi ai , a · b = ai bi , of any two vectors including
the scalar squares ξ · ξ , a · a, etc.,
• the determinants such as [abξ ] = ij k ai bj ξk of any three of the vectors.
10.2 Qij as an Isotropic Tensor

In the case of the double correlation, Qij , three vectors are available for the con-
struction of the invariant (10.3), i.e., a, a , and ξ . Then the fundamental invariants
are
a · a, a · ξ, a · ξ , [aa ξ ], a · a, a · a, ξ · ξ.
Since Qij ai aj
is a bilinear form in a and a,
the squares of quantities |a| and |a |,
etc., will not enter. We write b for a for consistency with the literature and then
have
Qij ai bj = Q1 (r)(a · ξ )(b · ξ ) + Q2 (r)(a · b) + Q3 (r)[abξ ]
= Q1 (r)ai bj ξi ξj + Q2 (r)ai bi + Q3 (r)ij k ai bj ξk . (10.5)
where r 2 = ξ · ξ . Since a and b are arbitrary, we must have
Qij = Q1 (r)ξi ξj + Q2 (r)δij + Q3 (r)ij k ξk . (10.6)
10.2 Qij as an Isotropic Tensor 39
The definition of isotropy requires that Qij ai bj be invariant under reflections

through P . Since Q3 εij k ξk changes sign under such reflections, isotropy requires
Q3 (r)ij k ξk = 0. (10.7)
Then,
Qij = Q1 (r)ξi ξj + Q2 (r)δij (10.8)
is an isotropic tensor.
10.2.1 Two More Examples
The Isotropic Vector
The problem is to find Li such that Li ai is invariant under the full rotation group,
where ai is an arbitrary unit vector. Only two vectors exist for this purpose: ai and ξi ,
so that the only available fundamental invariant is a · ξ . Hence,
Li ai = L(r)(a · ξ ) = L(r)ai ξi , (10.9)
and the isotropic vector is
Li = L(r)ξi . (10.10)
The Isotropic Tensor of Third Order
Consider the correlation of two components of u at P with a component of u at P .

The vectors a, b, and c = a are arbitrary in direction only, being of unit length.
Then we wish to find Tij k such that Tij k ai bj ck is invariant under the full rotation
group. We have seen that ij k ai bj ξk , etc., is not invariant under reflection through P ,
so we need not include such invariants in the present connection. Moreover, since
Tij k ai bj ck is trilinear, (a · a) will not enter either. Then we must have
Tij k ai bj ck = T1 (r)(a · ξ )(b · ξ )(c · ξ ) + T2 (r)(ξ · a)(b · c)

+ T3 (r)(b · ξ )(c · a) + T4 (r)(ξ · c)(a · b), (10.11)
or
Tij k ai bj ck = T1 ξi ξj ξk ai bj ck + T2 ξi ai bj cj + T3 ξj bj ai ci + T4 ξk ck ai bi . (10.12)
Since (10.12) must be true for arbitrary a, b, c, Tij k must be of the form
Tij k = T1 (r)ξi ξj ξk + T2 (r)ξi δj k + T3 (r)ξj δik + T4 (r)ξk δij . (10.13)
But we want to use Tij k as defined above, i.e.,
Tij k = ui uj uk , (10.14)

so that it must be symmetric in i and j . This requires that

T2 (r) = T3 (r),
and we have

Tij k = T1 (r)ξi ξj ξk + T2 (r) ξi δj k + ξj δik + T4 (r)ξk δij . (10.15)
10.3 Solenoidal Isotropic Tensors

The types of quantities we are interested in here are tensors up to order three which
are isotropic. The equation of continuity places the additional restriction on such
quantities that, when they are related to velocities, they should be solenoidal in at
least one index. We will next look briefly into the resulting properties of each of the
three orders of tensors.
10.3.1 Isotropic Vectors Li
In order that Li , an isotropic vector [see (10.10)], should be solenoidal, we must

have
∂Li ∂
= L(r)ξi = 0. (10.16)
∂ξi ∂ξi
Now,
∂ ∂r ξ
L(r) = L (r) = L (r) .
∂ξi ∂ξ r
Then,
ξi
L(r)δii + L (r) ξi = 0, (10.17)
r
or
3L(r) + rL (r) = 0. (10.18)
Equation (10.18) is equivalent to (r = 0)
∂ 3
(r L) = 0, (10.19)
∂r
which implies
r 3 L(r) = constant, (10.20)
or
const.
L(r) = . (10.21)
r3
In order that L(r) be continuous as r → 0, we require that the constant be zero,
whence
r 3 L(r) = 0. (10.22)
That is, there exists no non-vanishing, solenoidal isotropic vector.
10.3 Solenoidal Isotropic Tensors 41
Isotropic Second Order Tensor Qij
We have written Qij in terms of the two scalar functions Q1 and Q2 . Yet we know
that the transform ij of Qij has been expressed in terms of one scalar function,
the energy spectrum. However, our derivation of F (k) employed the equation of
continuity in the form uk ·k = 0. Therefore, we may expect that, if we now introduce
the equation of continuity, we may reduce the number of scalar functions needed to
specify Qij .
For the incompressible fluid, the continuity equation is
∂ui
div u = = 0. (10.23)
∂xi
If we operate div on Qij = ui uj at P , uj is not affected by the operation and we
have
∂Qij ∂ui uj

= = 0. (10.24)
∂xi ∂xi
Now,
ξi = xi − xi , (10.25)
so that
∂ ∂
Qij = − Qij , (10.26)
∂xi ∂ξi
and
∂
Qij = 0. (10.27)
∂ξi
Since Qij is symmetric in i and j , it is solenoidal in both indices. From (10.27) will
arise a relation between Q1 and Q2 .
If we substitute (10.8) into (10.27), we obtain
∂ Q2 (r)
Q1 (r)ξi ξj + Q2 (r)δij = Q1 (r)rξj + 3Q1 (r)ξj + Q1 (r)ξj + ξj
∂ξi r

Q2 (r)
= ξj Q1 (r)r + 4Q1 (r) +
r
= 0.
Or, since ξj is not in general zero,
1
Q1 r + 4Q1 + Q2 = 0. (10.28)
r
This gives the relation between Q1 and Q2 on whose account we may conclude that
Qij can be expressed in terms of one defining scalar.
10.3.2 Further Manipulations
If we take ∂Tij k /∂xk , we get an expression for a second order tensor, with sym-
metry in the indices. There result two equations, so that for a solenoidal isotropic
tensor, only one independent defining scalar is required. We shall return to this mat-
ter presently. But first we elaborate on the nature of the second order tensor.
As we have seen, the isotropic tensors may be represented by a defining scalar,
and so it is clear that the tensor equations governing isotropic turbulence may be
transformed into scalar equations governing these defining scalars. The immediate
problem is to pass to the scalar equations from the tensor equations. This requires
a direct way of specifying the defining scalar. This may be done most directly for
solenoidal tensors by introducing a ‘tensor potential’. That is, one can express a
tensor solenoidal in the index j by the curl of a tensor with respect to that in-
dex. Clearly, this potential tensor must be skew-symmetric, as its curl must be an
isotropic tensor. In the case of first order tensors, we have seen that there exist no
non-vanishing isotropic vectors. We go then to second order tensors.
If qij is a skew tensor, the isotropic, solenoidal tensor is
Qij = curl qij , (10.29)
where qij must also be skew-isotropic. The requirement that that qij ai bj be skew-
invariant, where ai and bj are arbitrary unit vectors is fulfilled when
qij ai bj = Q(r)ij k ai bj ξk . (10.30)
Here Q(r) is some arbitrary function of r and ij k ai bj ξk is the only determinant
from which qij ai bj can be constructed, which is consistent with the bilinear form
of qij ai bj . From the foregoing we may conclude that qij is of the form
qij = Q(r)ij k ξk . (10.31)
It follows from (10.29) that
∂qim
Qij = j lm . (10.32)
∂ξl
Qij is symmetric in i and j , and so must be solenoidal in i as well as j , although
the curl is taken in j .
∂
Qij = j lm (Qimk ξk )
∂ξl
∂
= −mj l mik (Qξk )
∂ξl
∂
= (δj k δli − δj i δlk ) (Qξk )
∂ξl
∂ ∂
= (Qξj ) − δij (Qξk )
∂ξi ∂ξk
Q
= ξi ξj + Qδij − 3Qδij − rQ δij .
r
Thus
Q
Qij = ξi ξj − (rQ + 2Q)δij , (10.33)
r
is the most general form of a solenoidal isotropic tensor of second order and Q is
the defining scalar of Qij . If we let
Q
Q1 = (10.34)
r
and
Q2 = −(rQ + 2Q), (10.35)
we see that (10.33) is identical with (10.8). Moreover, Q1 and Q2 as given in (10.34)
and (10.35) are seen to satisfy (10.28), the condition imposed by the equation of
continuity.
It may be further seen that the representation of Qij in terms of Q is unique. For,
when Qij = 0, the coefficients in (10.8) vanish and Q must vanish. Also, if Q = 0,
Qij = 0. In other words,
Q=0 ⇐⇒ Qij ≡ 0. (10.36)
We are now in a position to see how to pass from a tensor equation to an equation
in the defining scalar. Consider, for example, the equation
∂Qij
∇ 2 Qij = . (10.37)
∂t
Now

∇ 2 Qij = ∇ 2 curl(j ) qij = curl(j ) ∇ 2 qij , (10.38)
and our attention is diverted to ∇2q ij . We readily find this latter quantity to be given
by
∂2
∇ 2 qij = 2
Qij k ξk
∂ξl

∂ Q
= ξl ij k ξk + Qij l
∂ξl r

Q Q Q Q
= rij k ξk + 3 ij k ξk + ξl ij l + ij l ξl
r r r r

Q Q
= r +5 ij k ξk .
r r
That is,

Q
∇ 2 qij = Q + 4 ij k ξk . (10.39)
r
Since ∇ 2 qij is the skew-isotropic tensor from which we construct ∇ 2 Qij , the defin-
ing scalar of ∇ 2 Qij is given by
Q (r)
Q (r) + 4,
r
where Q(r) is the defining scalar of Qij . Then the tensor equation (10.37) passes
over into
∂ 2 Q 4 ∂Q ∂Q
+ = . (10.40)
∂r 2 r ∂r ∂t
The left-hand side of (10.40) is the radial component of the 5-dimensional Lapla-
cian. The strength of this method lies in the uniqueness of the relation between Q
and Qij .
10.3.3 The Isotropic Third Order Tensor, Tij k
We may now see how to find the defining scalar for Tij k . Originally, we wrote Tij k
in terms of 4 scalars [see (10.13)] but we will see how, for the special cases we are
interested in, one scalar will suffice to express Tij k . We set out as before to express
Tij k as the curl of a skew-tensor. Here Tij k is solenoidal in k, and we write it
∂
Tij k = klm tij m , (10.41)
∂ξl
where tij m is skew-isotropic. In order that tij m ai bj ck be skew-invariant, we must
express it in the appropriate combinations of determinants. (A product of two deter-
minants can always be expressed in terms of scalar products.) The available deter-
minants are [abc], [abξ ], [bcξ ], and [caξ ]. Now, tij k ai bj ck must be trilinear, so that
it will be a linear combination of [abc], [abξ ](c · ξ ), [bcξ ](a · ξ ), and [acξ ](b · ξ ).
In component form, these invariants are, respectively, to be expressed in terms of
the tensors ij k , ξk ij k ξl , ξi j kl ξl , and ξi kil ξl . In other words, Tij k must be a linear
combination of these tensors. This again would imply four fundamental scalars, if
the four tensors were independent. However, these tensors are not independent, as
we shall now see.
Consider the fourth order tensor
ij k ξl .
Among the results of permuting the indices, only 4 possible permutations could
possibly be independent because of the properties of ij k . These four are the ones
(any of many) with the factors ξi , ξj , ξk , or ξl . Since only three possible values of
ξi exist, it seems that even these four may not be independent. We then ask whether
it is possible to write an expression of the form
aij k ξl + blj k ξi + cilk ξj + dij l ξl = 0, (10.42)
that is, whether we may discover appropriate a, b, c, d such that (10.42) holds. We
may distinguish several different cases:
• i = j = k. In this case, (10.42) is identically true for all a, b, c, d.

• Two of i, j, k equal, and only two equal, e.g., i = j = k. Two possible subcases
arise:
1. l = i = j , which satisfies (10.42) for all a, b, c, d.
2. l =
i = j , which admits the two possible subcases: (i) l = k which is satisfied
by all a, b, c, d; (ii) l = k which is satisfied only if b = c.
Similarly one finds, by considering the cases i = k and j = k, that b = d and
c = d. Thus (10.42) can be true only if b = c = d.
• i, j, k all different. Then l must be the same as one of them. We will consider
l = i. Then
aij k ξl + blj k ξi = 0.
Since i = l, ξl = ξi and, in a loose way of writing,
aij k + bij k = 0,
which requires a = −b. It follows that, if
a = −b = −c = −d,
equation (10.42) is true. In particular, we may choose a = 1 and write
ij k ξl = lj k ξi + ilk ξj + ij l ξk
= ξi j kl + ξj kil + ξk ij l . (10.43)

If we multiply (10.43) by ξl , we find
r 2 ij k = ξi j kl ξl + ξj kil ξl + ξk ij l ξl . (10.44)
Hence, tij k can depend on at most three independent scalars.
Another restriction may be placed on tij k to make it gauge invariant. This latter
one is essentially a matter of convention. From (10.44), we see that the form of tij k
will be a linear function of tensors of the sort
Q(r) = ij k ξk ξl
and, as matters stand, Tij k will be unchanged if a gradient is added to this tensor.
Such gradients will have the appropriate form
∂ Q
Qij l ξl = ξk ij l ξl + Qij k . (10.45)
∂ξk r
In other words, Tij k is unchanged if (10.45) is added to tij k . However, the form of
the defining scalar will be affected and so we must establish some condition to make
tij k gauge invariant. The right-hand side of (10.45) is a linear combination of tensors
of the forms ij k and ξk ij l ξl , and rearrangement of (10.45) shows that they differ
by the gradient of a tensor. Thus Tij k is equivalently represented by
Q
− ij l ξl ξk and Qij k ,
r
and we will take them equal. That is,

Q
Qij k + ξk ij l ξl = 0. (10.46)
r
So, with the relations (10.46) and (10.44), we find that only two of ij k , ξk ij l ξl ,
ξi j kl ξl , and ξj kil ξl are independent, and we may completely specify tij k in terms
of any two of them. We choose to employ ξj ikm ξm and ξi j km ξm . Then,
tij k = T1 ξi j km ξm + T2 ξj ikm ξm . (10.47)
Now Tij k = ui uj uk is symmetric in i and j , and so tij k must be. This condition
implies
T1 = T2 = T (say), (10.48)
and

tij k = T ξi j km ξm + ξj ikm ξm , (10.49)
where T = T (r) is said to be the defining scalar of Tij k . To find Tij k , let us take the
curl of the first term on the right of (10.49). We have
∂ ∂
klm T ξi j ms ξs = klm j ms T ξi ξs
∂ξl ∂ξl
∂
= (δks δlj − δkj δls ) T ξi ξs
∂ξl
∂ ∂
= T ξi ξk − δkj T ξi ξl
∂ξj ∂ξl
T T
= ξj ξi ξk + T δij ξk + T δj k ξi − r 2 δj k ξi
r r
− 3T δkj ξi − T ξi δkj
T
= ξi ξk ξj − (3T + T r)ξi δj k + T δij ξk .
r
The second term from (10.49) is obtained by interchanging i and j . The result is
T (r)
Tij k = 2 ξi ξj ξk − T (r)r + 3T (r) (ξi δj k + ξj δik ) + 2T δij ξk . (10.50)
r
If Tij k ≡ 0, then T (r)/r = 0 and T = 0. Thus,
rT (r) + 3T (r) = 0 and 2T = 0,
whence we see that T = 0. Moreover, if T = 0, then Tij k ≡ 0. That is,
T =0 ⇐⇒ Tij k ≡ 0,
and the representation of Tij k by the defining scalar T (r) is unique.
Uniqueness Proof Let two tensors Uij k... and Vij k... be given by
∂
Uij k... = kmn q1 tnij ... (10.51)
∂xm
and
∂
Vij k... = kmn q2 tnij ... . (10.52)
∂xm
Then
∂
Uij k... − Vij k... = kmn (q1 − q2 )tnij ... , (10.53)
∂xm
so that the defining scalar of
Dij k... = Uij k... − Vij k...
is q1 − q2 . Then
q1 − q2 = 0 ⇐⇒ Uij k... − Vij k... ≡ 0
so that Uij k... ≡ Vij k... . We may suppose q1 = q2 = q (say). But Uij k... and Vij k...
cannot have the same defining scalar and be different. Moreover, if Uij k... ≡ Vij k... ,
then q1 = q2 . Finally, q1 = 0 ⇐⇒ Uij k... ≡ 0, and so if Vij k... and q2 are chosen ≡ 0
and = 0, respectively, the conditions on (10.51) and (10.53) become identical.
As an example of the passage to scalar equations from third order tensor equa-
tions, we consider
∂Tij k
∇ 2 Tij k = . (10.54)
∂t
The defining scalar of ∂Tij k /∂t is ∂T /∂t. The defining scalar of ∇ 2 Tij k is obtained
by finding ∇ 2 tij k (as in the case of ∇ 2 Qij ). We consider just one term in tij k and
get the rest of ∇ 2 tij k by making use of the symmetry:
∂2
∇ 2 T ξi j kl ξl = 2 (T ξi j kl ξl )
∂ξm

∂ T
= ξm ξi j kl ξl + T δim j kl ξl + T ξi j km
∂ξm r

T T T T
= rξi j kl ξl + 3 ξi j kl ξl + ξi j kl ξl + ξi j km ξm
r r r r
T T
+ ξi j kl ξl + T j ki + ξi j km ξm + T j ki
r r
T
= T ξi j kl ξl + 6 ξi j kl ξl + 2T j ki
r

T
= T + 6 ξi j kl ξl + 2T j ki .
r
Then,

T
∇ 2 ξj ikm ξm = T + 6 ξj ikl ξl + 2T ikj ,
r
and, on adding, we find

T

∇ tij k = T + 6
2
ξi j kl ξl + ξj ikl ξl . (10.55)
r
Thus, the defining scalar of ∇ 2 Tij k is T + 6T /r and this must be the same as the
defining scalar for ∂Qij /∂t. Then (10.54) passes over into
6 ∂T
T + T = . (10.56)
r ∂t
The left-hand side of (10.56) is the radial part of the 6-dimensional wave equation.
We will also be interested in the contracted tensor
∂Tikj
. (10.57)
∂ξk
Since this tensor is isotropic of second order, it must be symmetric in i and j . We
can get its defining scalar by performing a similar contraction on tikj . Thus
∂tikj ∂
= T ξi kj l ξl + T ξk ij l ξl
∂ξk ∂ξk
T
= ξk ξi kj l ξl + T ij l ξl + T rij l ξl + 3T ij l ξl + T ξk ij k .
r
Now,
ξk ξl kj l = −j kl ξk ξl = (ξ × ξ )j = 0.
Hence,
∂tikj
= (rT + 5T )ij l ξl ,
∂ξk
and the defining scalar of (10.57) is
rT + 5T . (10.58)
Chapter 11
The Karman-Howarth Equations
We may now proceed to the (statistical) dynamical aspects of isotropic turbulence

by introducing the equations of motion:
∂ui ∂ ∂
+ (ui uk ) = − + ν∇ 2 ui . (11.1)
∂t ∂xk ∂xi
If we multiply (11.1) by uj and average, we have
∂ui ∂
uj + ui uk uj = ν∇ 2 ui uj , (11.2)
∂t ∂xk
on observing that uj is a solenoidal isotropic vector and must vanish. Since
ξi = xi − xi , (11.3)
we have
∂ ∂ ∂ ∂
=− , = . (11.4)
∂xi ∂ξi ∂xi ∂ξi
Then (11.2) becomes
∂ui ∂
uj − Tij k = ν∇ 2 Qij . (11.5)
∂t ∂ξk
We may also express (11.1) in uj . It is
∂uj ∂ ∂
+ u u = − + ν∇ 2 uj . (11.6)
∂t ∂xk j k ∂xi
By multiplying (11.6) by ui , averaging and introducing (11.5), we obtain
∂uj ∂
ui + ui uj uk = ν∇ 2 ui uj . (11.7)
∂t ∂ξk
Recall from (10.50) that
T (r)
Tij k = 2 ξi ξj ξk − T r + 3T ξi δj k + ξj δik + 2T δij ξk .
r
50 11 The Karman-Howarth Equations
This expression is odd in ξi . Thus the correlation with r as origin with the point at
r + ξ is the negative of the correlation between these two points when r + ξ is taken
as origin. Therefore,
ui uj uk = −ui uj uk (11.8)
and
Tj ki = −uj uk ui . (11.9)
Then with (11.9), equation (11.7) becomes
∂uj ∂
ui + Tj ki = ν∇ 2 Qij . (11.10)
∂t ∂ξk
Adding (11.5) and (11.10), we have
∂ ∂
Qij = (Tj ki + Tikj ) + 2ν∇ 2 Qij .
∂t ∂ξk
Since ∂Tj ki /∂ξk is a second order isotropic tensor, it must be symmetric in i and j :
∂Qij ∂Tikj
=2 + 2ν∇ 2 Qij . (11.11)
∂t ∂ξk
By introducing the various defining scalars in these quantities, we find
2
∂Q ∂ ∂ 4 ∂
=2 r + 5 T + 2ν + Q, (11.12)
∂t ∂r ∂r 2 r ∂r
which is the well-known Karman–Howarth equation. We notice that the inertial term
gives rise to the appearance of the triple correlation. If the velocity distributions were
normal, the means of odd powers of the velocity would vanish. However, not only
does the triple correlation not vanish, but it is associated with the inertial term. We
may conclude that non-normality is fundamental to the turbulence phenomenon.
Chapter 12
The Meanings of the Defining Scalars
Consider the longitudinal correlation u u (see Fig. 12.1 left). In terms of (10.33),
with i = j ,
u u = rQ − (2Q + rQ )

= −2Q(r)
= f (r) in the older literature. (12.1)
The lateral correlation u⊥ u⊥ may also be found from (10.33) (see Fig. 12.1 right):
u⊥ u⊥ = −(rQ + 2Q)

= g(r). (12.2)
Accordingly,
1 ∂f (r)
g(r) = f (r) + r . (12.3)
2 ∂r
Equation (12.3) has been verified experimentally, thus strengthening confidence in
the concept of isotropy and in the use of the equation of continuity.
Now, Q(r) is an even function of r. This follows from the fact that
Qij (−r) = Qj i (r) = Qij (r)
and (10.33). Hence we may write
Q(r) = Q(0) + Q2 r 2 + · · · . (12.4)
Moreover, if we let r → 0 in (12.1) and (12.2), we find
u2 = −2Q(0) = f (0), (12.5)
u2⊥ = −2Q(0) = g(0), (12.6)

so that
u2 = u2⊥ (12.7)
52 12 The Meanings of the Defining Scalars
Fig. 12.1 Longitudinal and

Lateral correlations
Fig. 12.2 Longitudinal triple

correlation
and
1
Q(0) = − u2 . (12.8)
6
If there is a mean flow, we should consider (u − u )2 instead of u2 . Now
(u − u )2 = u2 + u2

− 2u u
= −2Q(0) − 2Q(0) + 4Q(r)

= 4 Q(r) − Q(0) . (12.9)
Next consider the longitudinal triple correlation u21 u1 (see Fig. 12.2). By (10.50),
u21 u1 = 2r 2 T − 2(rT + 3T )r + 2rT

= −4rT . (12.10)
Now we have seen that Tij k is odd in r, so that, according to (12.10), T (r) must be
even. Then we may express T (r) as a power series:
T (r) = T0 + r 2 T2 + · · · . (12.11)
An alternate means of expression for r small would be to expand u in a series
about P :

∂u1 ∂ 2 u1
u1 u1 = u1 u1 + r
2 2 +r 2 + ···
∂r ∂r 2
1 ∂ ∂ 2 u1
= u31 + r u31 + r 2 u21 2 + · · · . (12.12)
3 ∂r ∂r
If there is no mean motion, u31 = 0 and the first term in the series seems to be
∂ 2 u1
r 2 u21 .
∂r 2
However, we note that

∂ 2 ∂u1 ∂ 2 u1 ∂u1 2
u1 = u21 2 + 2u1 . (12.13)
∂r ∂r ∂r ∂r
12 The Meanings of the Defining Scalars 53
The term on the left is odd in ∂u1 /∂r and the second term on the right is odd to the
first power in u1 . We may conclude that
∂ 2 u1
u21 = 0.
∂r 2
It follows that the series (12.12) commences with the r 3 term. We may also combine
(12.10) with (12.11):
u21 u1 = −4rT (r)

= −4 rT0 + r 3 T2 + · · · . (12.14)
Now (12.14) must begin with a term in r3 because, as we have just seen, (12.12)
does. Thus, we must have
T0 = 0, (12.15)
whence (12.11) becomes
T (r) = r 2 T2 + · · · , (12.16)
or
T (r) = r 2 T2 + O(r 4 ) · · · . (12.17)
If there is a mean motion, we might consider
(u1 − u1 )3 = u31 − 3u21 u1 + 3u1 u2 − u3

1 1
2
= 3 u1 u1 − u1 u12
= −6u21 u1 = 24rT . (12.18)

Chapter 13
Some Results from the Karman–Howarth
Equation
13.1 The Taylor Microscale

To second order, according to (12.4) and (12.16), we have
Q = Q0 + Q2 r 2 (13.1)
and
T = r 2 T2 . (13.2)
We may substitute (13.1) and (13.2) into the Karman–Howarth equation:
∂Q0 ∂Q2
+ r2 = 4r 2 T2 + 10r 2 T2 + 16νQ2 + 4νQ2 .
∂t ∂t
This may be rearranged to give

∂Q0 ∂Q2
− 20νQ2 + − 14T2 r 2 = 0. (13.3)
∂t ∂t
The coefficients must vanish, and we have
∂Q0
= 20νQ2 . (13.4)
∂t
If we introduce (12.8), we find
∂Q0 1 d 2
=− u , (13.5)
∂t 6 dt
∂Q0 1
= ε = 20νQ2 , (13.6)
∂t 3
where ε is the rate of dissipation. Thus, it is clear why no constant term arose in
the series for T (r): the constant term gives rise to dissipation, while T arises in the
inertial term.
We may then write

Q2 2
Q(r) = Q0 1 + r + ··· . (13.7)
Q0
56 13 Some Results from the Karman–Howarth Equation
Now a correlation coefficient should equal 1 at r = 0, i.e., the correlation should go

as
r2
1− + ···,
λ2
where λ is the radius of curvature near the origin. λ is known as the microscale
(Taylor’s microscale) of turbulence.
From the foregoing, it is clear that Q0 /Q2 has the dimensions (length)2 . We may
readily see how the term microscale arises for λ. From (12.1) and (13.5), we have
∂f0
= 20νf2 , (13.8)
∂t
2
du21 u f2
= 20νf2 1 = 20νu21 , (13.9)
dt f0 f0
and
du21 20νu2
=− 2 1. (13.10)
dt λ
Multiplying by 3/2 and recalling that u2 = 3u21 , we see that (13.5) and (13.6) imply
that
15νu21
ε=− . (13.11)
λ2
Taylor verified (13.11) experimentally and found λ. Now λ is the size of the eddy
which causes the dissipation so it must represent the smallest eddies in the tur-
bulence. Thus it may also be concluded that viscosity causes dissipation and that
Stokes was correct on this point.
13.2 The Study of the Decay of Turbulence

In the stage of decay, the inertial term vanishes and the Karman–Howarth equation
reduces to the equation of diffusion:
2
∂Q ∂ Q 4 ∂Q
= 2ν + . (13.12)
∂t ∂r 2 r ∂r
Equation (13.12) admits the asymptotic solution
e−r /8νt
2
Q∼ . (13.13)
(νt)5/2
Because of the neglect of T , this is a normal function. The dissipating effect of
viscosity is again made clear. This solution (13.13) is the sort of thing that happens
when the viscosity dominates, i.e., when the Reynolds number is low. One then has
no effects from the inertial term, and if a random velocity is introduced, this sort of
laminar (noninteracting) turbulence can occur.
13.3 The Connection Between the Karman–Howarth Equation and the Kolmogorov Theory 57
But as far as our actual knowledge of turbulence as manifested through the iner-
tial term is concerned, these considerations have led only to the conclusion that the
concept of isotropy is a good one (at least as far as we have gone with the series for
f and g).
13.3 The Connection Between the Karman–Howarth

Equation and the Kolmogorov Theory
13.3.1 The Double Correlation
We saw that the fundamental length in Kolmogorov theory [see (8.3) on p. 30] is
3 1/4
ν
η= , (13.14)
ε
and that
[velocity] = (νε)1/4 . (13.15)
Then if f is a universal function,
1/4
ε
(u − u )2 = (νε) 1/2
f r 3 . (13.16)
ν
As R → ∞, (u − u )2 should become independent of ν. Thus, f must be of the
form c[r(ε/ν 3 )1/4 ]2/3 , where c is a constant. Then, if ν → 0,
1/4 2/3
ε
(u − u )2 = c(νε)1/2 r 3 = c(rε)2/3 . (13.17)
ν
It is clear that (13.17) is nearly true only when r is larger than λ, the Taylor mi-
croscale. Of course, as ν → 0, we have λ → 0.
Now we have by (12.9)

(u − u )2 = 4 Q(r) − Q(0) , (13.18)
so that, for ν → 0, r > λ, we have
Q(r) = Q(0) − cr 2/3 , (13.19)
where ε has been absorbed into c. Alternatively, if c is appropriately modified,

Q(r) = Q(0) 1 − cr 2/3 . (13.20)
Moreover we see that we must have

0 ≤
1 − cr 2/3
≤ 1, (13.21)
so that it is reasonable to take
−2/3
c = r0 ,
58 13 Some Results from the Karman–Howarth Equation
where r0 is the size of the largest eddies. Then,

2/3
r
Q(r) = Q(0) 1 − . (13.22)
r0
Now in the Kolmogorov theory (see Sect. 8.2)
vk ∼ k −1/3 ,
so
v ∼ r 1/3 ,
and
v 2 ∼ r 2/3 .
This seems consistent with (13.22).
13.3.2 The Triple Correlation
Here we expect (u1 − u1 )3 or u21 u1 to be of the form

1/4
ε
(νε)3/4 f r 3 .
ν
As ν → 0,
1/4
ε
(u1 − u1 )3 → (νε) 3/4
c r 3 = cεr. (13.23)
ν
With these considerations regarding the triple correlations, we may now enquire
what the Karman–Howarth equations imply in the case of stationary turbulence.
The first impulse might be to set ∂Q/∂t = 0. But this does not allow the energy to
be introduced into the equations. Instead, one notes that
∂Q ∂(Q − Q0 ) ∂Q0
= + . (13.24)
∂t ∂t ∂t
Then we assert, for stationary turbulence,
∂(Q − Q0 )
= 0. (13.25)
∂t
Also, according to (13.6),
∂Q0 1
= ε,
∂t 3
so that
∂Q 1
= ε (stationary turbulence). (13.26)
∂t 3
13.3 The Karman–Howarth Equation and the Kolmogorov Theory 59
Then the Karman–Howarth equation (11.12) becomes

2
1 ∂T ∂ Q 4 ∂Q
ε=2 r + 5T + 2ν + . (13.27)
3 ∂r ∂r 2 r ∂r
Upon multiplication of (13.27) by r 4 , we find

ε ∂ ∂ 4 d(Q − Q0 )
r 4 = 2 (r 5 T ) + 2ν r . (13.28)
3 ∂r ∂r dr
The operator ∂/∂r is equivalent to d/dr since all quantities depend upon r alone.
Hence integration of (13.28) is a straightforward matter. We have
ε 5 d
r = 2r 5 T + 2νr 4 (Q − Q0 ). (13.29)
15 dr
We can replace T and Q − Q0 according to (12.18) and (12.9):
ε 1 ν d
r = (u − u )3 + (u − u )2 . (13.30)
30 24 4 dr
We have seen that Q − Q0 ∼ r 2/3 . Hence,
d 2
(Q − Q0 ) ∼ r −1/3 ,
dr 3
which tends to zero as r → ∞. Hence,
4
(u − u )3 = εr, (13.31)
5
which agrees with the result from Kolmogorov theory and indicates the value of the
coefficient.
Chapter 14
The Relation Between the Fourth and Second
Order Correlations When the Velocity Follows
a Gaussian Distribution
Stewart has measured the quantity

(u − u)4
, (14.1)
(u − u)2
and found the value 2.9. The value for a Gaussian distribution is 3.0. This result
suggests that the fourth order moments of the velocities at two points are related to
the second order moments in the same way as for a normal distribution, except for
small values of r. Under these circumstances,
ui uj ul um = ui uj ul um + ui ul uj um + ui um uj ul . (14.2)
We now develop this relation.
14.1 Some Properties of the Gaussian Distribution
14.1.1 One-Dimensional Gaussian Distribution
For a one-dimensional distribution,

1
e−(y−a) /2σ ,
2 2
W (y) = √ (14.3)
2πσ
the mean occurs at y = a. If we measure y from its mean,
1
e−y
2 /2σ 2
W (y) = √ . (14.4)
2πσ
Now,
+∞ ∞
2
e−y
2 /2σ 2
W (y) dy = √ dy,
−∞ σ 2π 0
√
and on setting x = y/ 2σ , we obtain
62 14 The Relation Between the Fourth and Second Order Correlations
+∞ √ ∞ −x 2
2
W (y) dy = √ σ 2 e dx
−∞ σ 2π 0

23/2 1√
= 1/2 1/2 π
2 π 2
= 1.
So W is normalized. Further
∞
2
e−y /2σ y 2 dy
2 2
y2 = √
σ 2π 0

25/2 σ 3 ∞ −x 2 2
= √ e x dx
σ 2π 0
25/2 σ 3 1 √
= √ π = σ 2,
σ 2π 22
so σ is the dispersion.
Characteristic Function
The Fourier transform is

+∞
ψ(t) = e+ity W (y) dy. (14.5)
−∞
For the one-dimensional Gaussian distribution

+∞
1 2 2
ψ(t) = √ eity−y /2σ dy, (14.6)
σ 2π −∞
or
+∞
1 1
ψ(t) = √ exp − 2 (y − 2σ ity) dy
2 2
σ 2π −∞ 2σ
+∞
1 1
= √ exp − 2 (y − σ 2 it)2 + σ 4 t 2 dy
σ 2π −∞ 2σ
−σ 2 t 2 /2 +∞
e 1 2
= √ exp − 2 y − σ 2 it dy. (14.7)
σ 2π −∞ 2σ
Thus,
ψ(t) = e−σ
2 t 2 /2
, (14.8)
and
∞
n
σ 2t 2 1
ψ(t) = (−1)n . (14.9)
2 n!
n=0
14.1 Some Properties of the Gaussian Distribution 63
The Importance of the Characteristic Function
First,
+∞
ψ(0) = W (y) dy = 1. (14.10)
−∞
Now an alternate derivation of the moment relations is possible:
+∞
1
W (y) = e−ity ψ(t) dt, (14.11)
2π −∞
+∞ +∞ +∞
1
W (y) dy = ψ(t) dt e−ity dy. (14.12)
−∞ 2π −∞ −∞
Now,
+∞
1
δ(t) = e−ity dy, (14.13)
2π −∞
so
+∞
W (y) dy = ψ(0) = 1,
−∞
thereby showing that W (y) is normalized.
For the nth moment,
+∞ +∞ +∞
1
W (y)y dy =
n
ψ(t) dt e−ity y n dy (14.14)
−∞ 2π −∞ −∞
+∞
= (−i)−n δ (n) (t)ψ(t) dt. (14.15)
−∞
Successive integrations by parts give
+∞
1 ∂ nψ
W (y)y dy = n
n
. (14.16)
−∞ i ∂t n t=0
We have therefore
+∞
∂ψ
=i W (y)y dy = iy,
∂t t=0 −∞
(14.17)
∂ψ
= iy1 ,
∂t t=0
and

∂ nψ
= in yn . (14.18)
∂t n t=0
Or, alternatively, the nth moment is given by

1 ∂ nψ
yn = n , (14.19)
i ∂t n t=0
2n 2n
∂ ψ n ∂ ψ
y2n = (−1)−n = (−1) . (14.20)
∂t 2n t=0 ∂t 2n t=0
Introducing (14.9), we find

2n ∞
∂ (−1)k σ 2 k 2k
y2n = (−1)n t ,
∂t 2n k! 2 t=0
k=0
2n ∞ k 2 k
∂ (−1) σ
t 2k
∂t 2n k! 2
k=0
∞
k
(−1)k σ 2
= t 2k−2n (2k)(2k − 1) · · · (2k − 2n + 1).
k! 2
k=0
Of course, all terms for which n > k will have vanished under differentiation, and
this series is equivalent to
∞
∂ 2n (−1)k σ 2 k 2 k−n
ψ= (t ) (2k)(2k − 1) · · · (2k − 2n + 1). (14.21)
∂t 2n k! 2
k=n
Upon evaluation of (14.20) for t = 0, we find that the only non-vanishing term is
that for which k = n. We thus have
2n
∂ (−1)n σ 2 n
ψ = (2n)! (14.22)
∂t 2n t=0 n! 2
and

(2n)! σ 2 n (2n)! 2n
yn = = σ . (14.23)
n! 2 n!2n
We may verify this result:
∞
2
e−y /2σ y 2n dy
2 2
y2n = √
σ 2π 0
√ ∞
2
e−x x 2n dx
2
= √ ( 2σ ) 2n+1
σ 2π 0
2 √ 1 · 3 · 5 · · · (2n − 1) √
= √ ( 2σ )2n π
π 2n+1
1 · 3 · 5 · · · (2n − 1)
= 2(2σ 2 )n
2n+1
(2n)! 2n
= σ ,
n!2n
which agrees with (14.23).
14.1.2 n-Dimensional Gaussian Function
We have

1 1 1
W (y1 , . . . , yn ) = √ exp − B kl k l ,
y y (14.24)
(2π)n/2 |B| 2|B|
14.1 Some Properties of the Gaussian Distribution 65
where B = (bkl ) is the cofactor of bkl , i.e., the minor with appropriate sign, |B| is
the determinant of (bkl ), with bkl = yk yl and Bkl a symmetric tensor.
Characteristic Function
The characteristic function is

1
ψ(t1 , . . . , tn ) = exp − bkl tk tl . (14.25)
2
To prove this we must evaluate the integral
+∞ +∞
1
··· eiyk tk ψ(t1 , . . . , tn ) dt1 . . . dtn
(2π)n −∞ −∞
+∞ +∞
1 1
= ··· exp iyk tk − bkl tk tl dt1 . . . dtn .
(2π)n −∞ −∞ 2
Let
i
tk = uk + Bkr yr . (14.26)
|B|
Then

1 Bkr 1 i
iyk tk − bkl tk tl = iyk uk − yk yr − bkl uk ul + Bkr yr ul
2 |B| 2 |B|

i 1
+ Bls ys uk − B B y
kr ls r sy
|B| |B|2
Bkr 1 i
= iyk uk − yk yr − bkl uk ul − bkl Bkr yr ul
|B| 2 2|B|
i Bls
− bkl Bls ys uk + bkl Bkr yr ys .
2|B| 2|B|2
But
bkl Bkr = |B|δlr ,
so continuing from above,
1 Bkr 1
iyk tk − bkl tk tl = iyk uk − yk yr − bkl uk ul
2 |B| 2
1 1 1
− iδlr yr ul − iδks ys uk + δlr Bls yr ys
2 2 2|B|
Bkr yk yr 1
= iyk uk − − bkl uk ul
|B| 2
i i 1
− yl ul − yk uk + Brs yr ys
2 2 2|B|
1 1
=− Bkr yk yr − bkl uk ul .
2|B| 2
Hence,
+∞ +∞
1 1 1
W= exp − Bkr yk yr ··· exp − bkl uk ul du1 . . . dun .
(2π)n 2|B| −∞ −∞ 2
Now, bringing bkl uk ul into the form λk u2k , we find (with λ1 · · · λn = |B|)
+∞ +∞
1 1 1
W= exp − B y
kr k ry · · · exp − λ u2
k k du1 . . . dun
(2π)n 2|B| −∞ −∞ 2

1 1 (2π)n
= exp − B kr yk y r
(2π)n 2|B| λ1 · · · λn
1
= √ e−Bkr yk yr /2|B| .
(2π)n/2 |B|
Let F (y1 , . . . , yn ) be a polynomial in (y1 , . . . , yn ). Then by Fourier’s integral theo-
rem,
+∞ +∞
F (y1 , . . . , yn ) = ··· W (y1 , . . . , yn )F (y1 , . . . , yn ) dy1 . . . dyn
−∞ −∞
+∞ +∞
1
= · · · F (y1 , . . . , yn ) dy1 . . . dyn
(2π)n −∞ −∞
+∞ +∞
× ··· eiyk tk ψ(t1 , . . . , tn ) dt1 . . . dtn
−∞ −∞
+∞ +∞
1
= ··· ψ(t1 , . . . , tn ) dt1 . . . dtn
(2π)n −∞ −∞
+∞ +∞
× ··· F (y1 , . . . , yn )eiyk tk dy1 . . . dyn
−∞ −∞

1 ∂ 1 ∂
= F ,..., ψ(t1 , . . . , tn ) . (14.27)
i ∂t1 i ∂tn t1 =···=tn =0
Hence,

∂2
yk yl = − ψ(t1 , . . . , tn ) .
∂tk ∂tl t1 =···=tn =0
Since

1
ψ(t1 , . . . , tn ) = exp − bkl tk tl ,
2
2
∂ 1
yk yl = − exp − brs tr ts .
∂tk ∂tl 2 t1 =···=tn =0
Now

∂ 1 1 1
exp − brs tr ts = − brs (ts δrl + tr δls ) exp − brs tr ts
∂tl 2 2 2
14.2 Addition Theorem for Gaussian Distributions 67

1 1
= − (bls ts + brl tr ) exp − brs tr ts
2 2

1
= −blm tm exp − brs tr ts ,
2
and

∂2 1 1
exp − brs tr ts = − blk exp − brs tr ts
∂tk ∂tl 2 2

1
− blm tm bkn tn exp − brs tr ts
2

1
= (−blk + blm bkn tm tn ) exp − brs tr ts .
2
Hence,
yk yl = blk = bkl .
14.1.3 Two-Dimensional Gaussian Function
In the two-dimensional case,

y12 y1 y2 σ2 ρσ τ
B= = ,
y1 y2 y22 ρσ τ τ2
where ρ is the correlation coefficient, i.e.,

y1 y2 y1 y2
ρ= = ,
2 2 στ
y1 y2
and
B11 = τ 2 , B12 = −ρσ τ, B22 = σ 2 , |B| = σ 2 τ 2 (1 − ρ 2 ).
Finally,
2
1 1 y1 y22 2ρy1 y2
W (y1 , y2 ) = exp − + − .
2πσ τ 1 − ρ 2 2(1 − ρ 2 ) σ 2 τ 2 στ
14.2 Addition Theorem for Gaussian Distributions
Let x1 , . . . , xn be distributed according to

1 1 1 xi2
W (x1 , . . . , xn ) = exp − .
(2π)n/2 σ1 σ2 · · · σn 2 σi2
Let y1 , . . . , ys (s ≤ m) be s linear combinations of the xi , i.e.,

yk = aki xi , k = 1, . . . , s,
where the aki are constants. The yk are then distributed according to an s-
dimensional distribution

1 1
W (y1 , . . . , ys ) = √ exp − Bkl yk yl ,
(2π)s/2 |B| 2|B|
where

n
bkl = yk yl = aki ali σi2 .
i=1
Proof Clearly,
+∞ +∞
s
W (y1 , . . . , ys ) = ··· W (x1 , . . . , xn ) δ(yk − aki xi ) dx1 . . . dxn
−∞ −∞ k=1
+∞ +∞
1 1 xi2
= ··· exp − dx1 . . . dxn
(2π)s+n/2 σ1 · · · σn −∞ −∞ 2 σi2
+∞ +∞
× ··· ei(tk yk −tk aki xi ) dt1 . . . dtn
−∞ −∞
+∞ +∞
1
= ··· eiyk tk dt1 . . . dts
(2π)s+n/2 σ1 · · · σn −∞ −∞
+∞ +∞
1 xi2
× ··· exp − − iaki tk xi dx1 . . . dxn .
−∞ −∞ 2 σi2
Now

1 xi2 1 1
+2 aki tk xi = xi + 2σi
2 2
aki tk xi
2
i
σi2 k
2
i
σi2 k
2
1 1
= x i + σi
2
a t
ki k
2
i
σi2 k

− σi4 aki tk ali tl
k l
1 2
= xi + σi2
aki tk
i
2σi2 k
1 2
− σi aki ali tk tl .
2
i k l
Let

bkl = aki ali σi2 .
i
14.3 Proof of (14.2) 69
Then
2
1 xi2 1 1
2
+2 aki tk xi = 2
xi + σ i2
aki tk − bkl tk tl ,
2 σi 2σi 2
i k i k
and

1
W (y1 , . . . , ys ) = ··· eiyk tk −bkl tk tl dt1 . . . dtn
(2π)s+n/2 σ1 · · · σn

1
× · · · exp − xi + σi a t
ki k dx1 . . . dxn ,
i
2σi2 k

1 1
W (y1 , . . . , yn ) = · · · exp iy t
k k − b kl k l dt1 . . . dtn ,
t t
(2π)s 2
as required.
14.3 Proof of (14.2)

Suppose that the six quantities ui and uj are distributed according to a six-
dimensional Gaussian distribution. The characteristic function must be of the form

1
exp − yk yl tk tl .
2
Let αi → ui , βj → uj . We have ui uj , ui uj , and ui uj , and

1
ψ(α, β) = exp − ui uj αi αj + 2ui uj αi βj + ui uj βi βj .
2
By the general theorem on moments, viz., (14.27),
4

1 ∂4
ui uj ul um = ψ(α, β) .
i ∂αi ∂αj ∂βl ∂βm α=β=0
Replace i and j in ψ(α, β) by s and t. Then
∂ 1
ψ(α, β) = − 2us um αs + us um βs + ut um βt ψ(α, β)
∂βm 2

= − us um + us um ψ(α, β),
∂ 2 ψ(α, β)
= us um αs + us um βs us ul αs + us ul βs − ul um ψ(α, β)
∂βl ∂βm
= Aψ(α, β),
3
∂ ψ
= − uj us αs + uj ut βt A + uj um us ul αs + us ul βs
∂αj ∂βl ∂βm

+ uj ul us um αs + us um βs ψ(α, β)
= Bψ(α, β),
and
∂ 4ψ
= − Aui uj − uj us αs + us ut βt ui um us ul αs + us ul βs
∂αj ∂αj ∂βl ∂βm

+ ui ul us um αs + us um βs

+ uj um ui ul + uj ul ui um − B us ui αs + ui us βs ψ(α, β).
When α = β = 0, we then find

ui uj ul um = ui uj ul um + ui um uj ul + ui ul uj um ,
which is (14.2).
Chapter 15
Chandrasekhar’s Theory of Turbulence
In the derivation of the Karman–Howarth equation, we multiplied the equation of

motion in terms of the velocity at xi by the velocity at xi . But in order to obtain
∂Qij /∂t, a similar procedure was followed in which the roles of these two veloci-
ties were interchanged. The results of these two operations, when combined, gave
∂Qij /∂t. If, however, the equation of motion had in the first place been multiplied
by a velocity at another time, say t , at xi , only one operation would have been
needed to obtain the desired equation in the correlations. Moreover, after Q(r, t)
has been introduced by requirements of convenience (t is the interval between t and
t ), it is seen that just such a description of turbulence is what is needed for a general
theory. For we have seen that, both physically and mathematically, one cannot get an
equation in Q(r) alone. More importantly, perhaps, Q(r) cannot include a descrip-
tion of the phase relationships that we have seen to be fundamental to a description
of turbulence as well as to a study of its dynamical properties.
Having made this change in approach, we pose ourselves the problem of getting
an equation in Q(r, t) alone. As an aside note that, from what has been said, it is
clear that Q(r, t) is the defining scalar of
Qij = ui (xi , t)uj (xi , t ) = ui uj . (15.1)
Now we might continue the process of multiplying the equation of motion by quan-
tities leading to the introduction of correlations. In particular, we might multiply by
ul um and introduce an additional equation relating the third and fourth order corre-
lations. But each new equation introduced in this way brings with it the introduction
of another defining scalar, unless some additional condition is imposed. In other
words, we can get a new equation without bringing a new incalculable scalar only if
the new scalar can be related back to some other known quantity. Such a condition
could be an assumption concerning the state of motion of the fluid. It might be as-
sumed that the velocity distribution is Gaussian. In this case the third moment, i.e.,
the triple correlation, would vanish. Though it is true that the triple correlations are
found to be small experimentally, their finiteness is fundamental to the turbulence
problem. However, the fourth order correlations behave as if they were nearly Gaus-
72 15 Chandrasekhar’s Theory of Turbulence
sian, and these might be calculated on the assumption of a Gaussian distribution in

accordance with (14.2). In this way, a closure to the number of scalars introduced
with each correlation might be achieved. Thus, in terms of the notation of correla-
tions, we could make (14.2) a hypothesis as follows:
2
Qij ;lm = u21 δij δlm + Qil Qj m + Qim Qj l . (15.2)
With this hypothesis and the introduction of concepts implicit in Q(r, t), we proceed
to attempt the derivation of an equation in Q alone.
We multiply
∂ui ∂ ∂
+ uk ui = − + ∇ 2 ui (15.3)
∂t ∂xk ∂xi
by uj = uj (xi , t ). Of course, ui uj is a solenoidal isotropic tensor, and (10.33)
applies, i.e.,
Q
Qij = ui uj = ξi ξj − (rQ + 2Q)δij . (15.4)
r
We will adopt the convention
t − t = τ > 0.
Then
∂ ∂
Qij − Tikj = ν∇ 2 Qij , (15.5)
∂τ ∂ξk
since
∂ ∂
uj = uj ,
∂xi ∂xi
and uj is a solenoidal isotropic vector and must vanish.
Now multiply (15.3) by ul um and average. We have
∂ ∂ ∂
− Tlmi − Qij ;lm = Plm − ν∇ 2 Tlmi , (15.6)
∂τ ∂ξj ∂ξi
where
Plm = ul um (15.7)
and we have used
ul um ui = −ul um ui = −Tlmi .
We shall now seek to pass from (15.5) and (15.6) into scalar equations. Let
∂2 n−1 ∂
Dn = 2
+ . (15.8)
∂r r ∂r
15 Chandrasekhar’s Theory of Turbulence 73
We have seen that the quantities in (15.7) have the following defining scalars:
∂ ∂
Qij → Q,
∂τ ∂τ
∂
Tikj → rT + 5T [see (10.58)] , (15.9)
∂ξk
∇ 2 Qij → D5 Q [see (10.39)] . (15.10)

Hence (15.5) passes over into

∂
− νD5 Q = rT + 5T . (15.11)
∂τ
We turn to (15.6). From (15.7), we see that Plm must be symmetric in its indices,
and hence, by (10.8),
Plm = P1 ξl ξm + P2 δlm . (15.12)
Unfortunately, Plm is not solenoidal. We may rewrite (15.6) in the form

∂ ∂ ∂
− ν∇ Tlmi = −
2
Plm + Qij ;lm . (15.13)
∂τ ∂ξi ∂ξj
In obtaining (15.13) from (15.3) as we had written it, no manipulations involving i
were performed. Thus, since (15.3) was solenoidal in i, (15.13) must be. Moreover,
the left-hand side of (15.13) is solenoidal in i. Then
∂ ∂
Plm + Qij ;lm
∂ξi ∂ξj
must also be solenoidal in i. This quantity is also symmetric in l and m, so that it
must have a single defining scalar, say X. We write
∂ ∂
Xlmi = Plm + Qij ;lm . (15.14)
∂ξi ∂ξj
Then (15.14) must pass over into

∂
− νD7 T = −X. (15.15)
∂τ
Now (15.11) is essentially the Karman–Howarth equation for Q(r, τ ), and (15.15)
relates the third and fourth order correlations. We now attempt to relate X and Q
by means of the statistical hypothesis (15.2). Clearly, an equation in Q and X in
addition to (15.11) and (15.15) will be sufficient to provide an equation in Q alone.
We have, from (15.2),
∂ ∂Qil ∂Qim
Qij ;lm = Qj m + Qj l . (15.16)
∂ξj ∂ξj ∂ξj
Now
Q
Qil = ξi ξl − (rQ + 2Q)δil
r
= Q1 ξi ξl + Q2 δil ,
and
∂Qil Q Q
= 1 ξi ξj ξl + Q1 δij ξl + δlj ξi + 2 δil ξj . (15.17)
∂ξj r r
Hence,

∂Qil Q Q2 Q2 Q2
Qj m = rQ1 Q1 + 2Q21 + 1 ξi ξl ξm + rQ1 Q2 + ξm δil
∂ξj r r
+ Q1 Q2 ξl δim + Q1 Q2 ξi δlm , (15.18)
and

∂ Q1 Q2
Qij ;lm = 2 rQ1 Q1 + 2Q1 +
2
ξi ξl ξm
∂ξj r

Q2 Q2
+ rQ1 Q2 + (ξm δil + ξl δim )
r
+ Q1 Q2 (ξm δil + ξl δim ) + 2Q1 Q2 ξi δlm . (15.19)
Now,
Q
Q1 =
r
and
Q2 = −(rQ + 2Q).
Hence,
Q Q
Q1 = − 2
r r
and
Q2 = −(rQ + 3Q ).
We have

1 Q2
rQ1 Q1 =
QQ − , (15.20)
r r
Q2
2Q21 = 2 , (15.21)
r2

Q1 Q2 1 QQ
= 2 Q2 + 2 − rQ Q − 2QQ , (15.22)
r r r
whence

Q2 QQ QQ
(15.20) + (15.21) + (15.22) = 2 − + .
r2 r2 r3
Further,

rQ1 Q2 = − rQ Q + 3Q2 , (15.23)
Q2 Q2 6
= QQ + 3Q Q + 2QQ + rQ Q , (15.24)
r r

2 2QQ
Q1 Q2 = − Q + , (15.25)
r
whence
4
(15.23) + (15.24) + (15.25) = 2QQ + QQ − Q2 .
r
Hence,
2
∂ Q QQ QQ
Qij ;lm = 4 − + ξi ξl ξm
∂ξj r2 r2 r3

4
+ 2QQ + QQ − Q2 (ξm δil + ξl δim )
r

2 QQ
−2 Q +2 ξi δlm . (15.26)
r
Now, from (15.12), we see that
∂ P P
Plm = 1 ξi ξl ξm + P1 (ξm δil + ξl δim ) + 2 ξi δlm . (15.27)
∂ξi r r
Combining (15.26) and (15.27), we may write an expression for Xlmi according to
(15.14). This expression will be equivalent to
Xlmi = T1 ξi ξl ξm + T2 (ξm δil + ξl δim ) + T4 ξi δlm , (15.28)

since Xlmi is a third order isotropic tensor, symmetric in l and m and solenoidal in
i [see (10.50)]. We may associate coefficients of the various tensors. We find
2 ⎫
Q QQ QQ P1 ⎪
T1 = 4 − + + , ⎪
⎪
r2 r2 r3 r ⎪ ⎪
⎪
⎪
⎬
4 2
T2 = 2QQ + QQ − Q + P1 , (15.29)
r ⎪
⎪
⎪
⎪
P QQ ⎪
⎪
T4 = 2 − 2 Q2 + 2 . ⎪
⎭
r r
Inspection of (10.50) shows that, if X is the defining scalar of Xlmi , then
⎫
2 ⎪
T1 = X , ⎪
⎪
r ⎬
T2 = −(rX + 3X), ⎪ (15.30)
⎪
⎪
⎭
T4 = 2X.
In order to relate P1 and P2 , we make use of the knowledge that Xlmi is solenoidal
in i. Thus,

∂Xlmi 2T
= rT1 + 5T1 + 2 ξl ξm + (2T2 + 3T4 + rT4 )δlm
∂ξi r
≡ 0, (15.31)
and we must have

⎫
2
rT1 + 5T1 + T2 = 0, ⎬
r (15.32)
⎭
2T2 + rT4 + 3T4 = 0.
We introduce (15.29) and find, after noting that

QQ QQ Q2 Q Q QQ P1 P1
T1 = 4 + 3 − − − 3 + − 2,
r2 r3 r3 r2 r4 r r
4 4 4
T2 = 2QQ + Q2 + QQ − 2 QQ + P1 ,
r r r

P P 2Q2 2QQ 2QQ
T4 = 2 − 22 − 2 2Q Q + + − ,
r r r r r2
that the conditions (15.32) become
⎫
6 QQ Q2 ⎪
P1 + P1 = −4 +6 2 , ⎪
⎬
r r r
(15.33)
2 ⎪
⎪
P2 + P2 + 2P1 = 4(3Q2 + rQ Q ). ⎭
r
We might combine these two foregoing equations, but elect instead to follow an
alternate procedure. We note that, if we contract Xlmi in l and m, the resulting Xlli
is an isotropic, solenoidal vector and must vanish. That is,
Xlli ≡ 0.
This gives us
2T2 + rT4 + 3T4 = 0, (15.34)
which clearly must be a linear combination of the two equations (15.33). If we
substitute (15.29) into (15.33), we find
3
P + rP1 + 2P1 = 4Q2 . (15.35)
r 2
We may eliminate P1 between (15.35) and the second of (15.33):
P2 1 8
− 2 P2 − P1 = Q2 + 4Q Q . (15.36)
r r r
This equation may also be written

∂ P2 2
− P1 = 4 Q Q + Q2 . (15.37)
∂r r r
From (15.30), we note that
−(5X + rX ) = T2 − T4

QQ P2
= 2QQ + 8 + Q2 − − P1 ,
r r
and, upon introduction of (15.37), we find

∂ 4 4
− (5X + rX ) = 2Q Q + Q − 2 Q , (15.38)
∂r r r
which may be rewritten

∂ ∂ ∂ 2 Q 4 ∂Q
− (5X + rX ) = 2Q + . (15.39)
∂r ∂r ∂r 2 r ∂r
Thus (15.38) together with (15.11) and (15.15) form a system of equations from
which may be found a single equation in Q alone. The system is

∂
− νD5 Q = rT + 5T , (15.40)
∂τ

∂
− νD7 T = −X, (15.41)
∂τ
∂ ∂
− (5X + rX ) = 2Q D5 Q. (15.42)
∂r ∂r
It is possible to eliminate T from these equations in a simple way. We note that

∂ ∂
5+r D7 = D5 5 + r . (15.43)
∂r ∂r
We may easily verify (15.43). The left-hand side when expanded is

∂ ∂2 6 ∂ ∂2 30 ∂ ∂3 ∂2 6 ∂
5+r + =5 + + r + 6 −
∂r ∂r 2 r ∂r ∂r 2 r ∂r ∂r 3 ∂r 2 r ∂r
∂3 ∂2 24 ∂
=r + 11 + ,
∂r 3 ∂r 2 r ∂r
while the right-hand side becomes

∂2 4 ∂ ∂ ∂2 20 ∂ ∂2 ∂3 4 ∂ ∂2
+ 5+r =5 + +2 2 +r 3 + +4 2
∂r 2 r ∂r ∂r ∂r 2 r ∂r ∂r ∂r r ∂r ∂r
∂3 ∂2 24 ∂
=r 3
+ 11 2 + ,
∂r ∂r r ∂r
and (15.43) is verified.

In the light of the above identity, we rewrite (15.41), (15.40) and (15.42) as

∂ ∂
− νD5 Q = 5 + r T, (15.44)
∂τ ∂r

∂
− νD7 Q = −X, (15.45)
∂τ

∂ ∂ ∂
5+r X = −2Q D5 Q. (15.46)
∂r ∂r ∂r
First, we operate (∂/∂τ − νD5 ) on (15.44), and introduce (15.45) and (15.46):
2
∂ ∂ ∂
− νD5 Q= − νD5 5+r T
∂τ ∂τ ∂r

∂ ∂
= 5+r − νD5 T
∂r ∂τ

∂
=− 5+r X, (15.47)
∂r
2
∂ ∂ ∂ ∂
− νD5 Q=− 5+r X
∂r ∂τ ∂r ∂r
∂
= 2Q
D5 Q. (15.48)
∂r
We have thus seen how an equation in Q alone may be derived, subject to the
statistical hypothesis made.
Chapter 16
A More Subjective Approach to the Derivation
of Chandrasekhar’s Equation
The equation in Q(r, t) alone was derived on the basis of three equations, namely
(15.44), (15.45) and (15.46). However, the derivation of (15.46) in the form given
clearly anticipated the knowledge that a simple elimination of X and T would be
possible. This chapter is devoted to showing how (15.46) really arises following a
more straightforward line of reasoning.
We have, in the absence of viscosity,
∂ui ∂ ∂
+ ui uk = − . (16.1)
∂t ∂xk ∂xi
The usual procedure would be to multiply by uj = uj (xi ) and average. For the new
theory, we multiply by uj (t ) and average. We find
∂ ∂ ∂
uj (t )ui (t) = − |uj (t )ui (t)uk (t)| − (t)uj (t ). (16.2)
∂t ∂xk ∂xi
We note that, if T < t ,
T
∂
uj (t ) = −
u (t ) dt + uj (T ). (16.3)
∂t j t
However, we have from the equations of motion,
∂ ∂ ∂
uj = − ui uk − , (16.4)
∂t ∂xk ∂xi
and so
T ∂ ∂ (t )

uj (t ) = u (t )u (t ) + dt + uj (T ). (16.5)
t ∂xl j l
∂xj
If we substitute (16.5) into the right-hand side of (16.2), we find, if T → ∞,
∞
∂ ∂ ∂ ∂ (t )
Qij = uj (t )ul (t ) + ui (t)uk (t) dt
∂τ t ∂ξk ∂ξl ∂ξl
∂
+ lim u (T )uk (t)ui (t), (16.6)
T →∞ ∂ξk j
82 16 A More Subjective Approach to the Derivation of Chandrasekhar’s Equation
where τ = t − t and ξi = xi − xi .

The term in (t)uj t vanishes. As T → ∞, τ → ∞, and the triple correlation
vanishes. Then the last term on the right of (16.6) vanishes and (16.6) becomes
∞
∂ ∂ ∂ ∂
Qij = Qik;j l + Pik dτ. (16.7)
∂τ τ ∂ξk ∂ξl ∂ξj
Now we saw that
∂ ∂
Xikj = Qik;j l + Pik (16.8)
∂ξl ∂ξl
is symmetric in i and k, and solenoidal in j . Hence, according to (10.58), the defin-
ing scalar of ∂Xikj /∂ξk is
rX + 5X.
Then if we pass to the scalar equation of (16.7), we have
∞
∂Q
= (5X + rX ) dτ, (16.9)
∂τ τ
and
∂ 2Q
= −(5X + rX ). (16.10)
∂τ 2
This is the motivation behind writing the expressions leading to (15.38), and in
deriving (15.42). Once we obtain (15.42), we look for an identity such as (15.43)
and proceed in a straightforward way.
The simplifications which appeared in the elimination of X and T are in a gen-
eral way both surprising and encouraging. One could not have expected such simple
mathematics, unless the mathematics is a good representation of elementary physi-
cal ideas. In this sense, the theory justifies some attention.
Chapter 17
The Dimensionless Form of Chandrasekhar’s
Equation
We introduce
u u
f (r, t) = , (17.1)
u2
and, by (12.1), write
u2 f (r, t) = −2Q(r, t). (17.2)

Also, we will use as the unit of length and

u2
as the unit of time. Then, using (17.2), we may rewrite (15.48) in dimensionless
fashion, introducing the new units. We note that the dimensions of ν are
L2
[ν] =
T
and that

∂2 5 ∂ 1
[D5 ] = + = 2.
∂r 2 r ∂r L
We wish
to keep ν in evidence so, to make it dimensionless, we replace it by
ν/( u21 ), which is the inverse of a Reynolds number based on these units. Further,
we bring D5 to dimensionless form by replacing it with 2 D5 . By similar manipu-
lations with the remaining operators, we find the dimensionless form of (15.48):
2
∂ ∂ ∂
− νD5 f = −f D5 f. (17.3)
∂r ∂t ∂r
Chapter 18
Some Aspects and Advantages of the New
Theory
The Karman–Howarth equation relates the two scalars Q and T . For a deductive
theory of turbulence, we would like to find a way of isolating an equation in one of
these scalars. Although the possibility of introducing a normal velocity distribution
is appealing, it has the effect of requiring
T = 0,
and thereby demolishing the effects of the inertial term. Though T is small, it is
definitely nonzero in fully developed turbulence.
Thus, we were led to calculating T , a small quantity, by considering that the
fourth order correlations behave as though the velocity is normally distributed. This
is essentially the calculation implied by (16.5). For if we multiply both sides of
(16.5) by ui ul and average, we obtain a relation between the third and fourth order
correlations by letting T → ∞.
There are at least two advantages of the new theory. These are:
• a mathematical justification of the assumptions of the Heisenberg theory,
• compatibility with the Kolmogorov theory.
18.1 A Mathematical Justification of the Assumptions

of the Heisenberg Theory
Equation (15.48), the equation for Q, is
2
∂ ∂ ∂
− νD5 Q = 2Q D5 Q,
∂r ∂t ∂r
and for ν = 0, it becomes
∂3 ∂
2
Q = 2Q D5 Q. (18.1)
∂t ∂r ∂r
86 18 Some Aspects and Advantages of the New Theory
When the inertial term is neglected, we have

∂ui ∂
=− + ν∇ 2 ui ,
∂t ∂xi
∂ui uj
= ν∇ 2 ui uj ,
∂t
or
∂Q
= νD5 Q. (18.2)
∂t
This becomes
∂ 2Q ∂
= ν D5 Q, (18.3)
∂r∂t ∂r
and we see that the inertial and viscous terms differ in their effects by a time deriva-
tive.
Expanding (15.48), we have
2
∂ ∂ ∂ ∂
2
+ 2νD5 + ν D5 Q = 2Q D5 Q.
2 2
(18.4)
∂r ∂t ∂t ∂r
If we neglect O(ν 2 ),

∂3 ∂ ∂
2
Q = 2 Q − ν D5 Q. (18.5)
∂r∂t ∂t ∂r
We see that the operators −Q and ν∂/∂t are the effects, respectively, of the inertial
term [see (18.1)] and of the viscous term [see (18.3)], and that they combine linearly.
This similarity in the behaviour of the two operators is the essential assumption of
the Heisenberg theory.
18.2 Compatibility with the Kolmogorov Theory

We choose our scale so that
f (0) = 1.
The Kolmogorov theory tells us that in this case [see (13.20)]
2/3
r
f (r) = 1 − .
r0
In the Kolmogorov case of ν → 0, (17.3) becomes
∂ 3f ∂
2
= −f D5 f. (18.6)
∂t ∂r ∂r
18.2 Compatibility with the Kolmogorov Theory 87
If we consider only values of f not far from unity, and substitute
f =1−g (18.7)
in the foregoing, we may neglect O(g 2 ). We then have
∂ 3g ∂
2
= − D5 g. (18.8)
∂r∂t ∂r
Alternatively, we may write
∂ 2g
= −D5 g. (18.9)
∂t 2
Since at t = 0, we have g ∼ r 2/3 ,
2
∂ g
= −const. × D5 r 2/3 = −const. × r −4/3 . (18.10)
∂t 2 t=0
Equation (18.10) indicates a life time of r −2/3 for an eddy of size r. The Kol-
mogorov theory predicts a life time of
1
τ≈ . (18.11)
vk k
Now (see p. 29)
vk ∼ k −1/3 ,
so that
τ ∼ r −2/3 ,
in agreement with the result from (18.10). (This demonstration of compatibility is
due to Fermi.)
Chapter 19
The Problem of Introducing
the Boundary Conditions
Since we are considering isotropic turbulence, the presence of a boundary seems

inconsistent with the framework. In particular, it is difficult to see how one could
introduce ε into the problem at all.
If we neglect viscosity, we have

∂ 3f ∂ ∂ 2f 4 ∂f
= −f + . (19.1)
∂r∂t 2 ∂r ∂r 2 r ∂r
One boundary condition of which we may be certain is
f (0, 0) = 1. (19.2)
It is tempting to say that f should be bounded as r → ∞, but it is not feasible to

deal with r greater than the largest eddies present. We do have, though, the following
(apparently) reasonable condition:
f (r, t) → 0 as t → ∞ (for all r) . (19.3)
Finally, it may be possible to introduce ε if we include viscosity. For in that case,

i.e., ν = 0, we have, according to (13.11),
2
∂ f
15νu1 2 = −ε. (19.4)
∂r 2 0,0
But we refrain from applying this at the present state of the theory.
In the framework of the Kolmogorov theory, we would expect (18.11) to yield
only one solution (apart from scale factors), and indeed (18.11) does have certain ho-
mologous properties. But, more generally, one feels that a turbulent medium (apart
from intrinsic properties like ν) may exhibit states of motion differing by more than
scale transformations and differing according to the value of ε associated with the
medium. Accordingly, we may expect (18.11) to yield a one parameter family of
solutions, where different values of the parameter correspond to different values
of ε.
Chapter 20
Discussion of the Case of Negligible Inertial
Term
If we ignore the inertial term in the equation of motion, we find that the equation for
f is, according to (18.2),
2
∂f ∂ f 4 ∂f
=ν + . (20.1)
∂t ∂r 2 r ∂r
In this situation, we may use (19.4) as a boundary condition. Of course, (19.2) is
valid too. We have seen that we cannot discuss the behaviour of f as r → ∞, but
we can use as a restriction some condition, such as r tends to some r0 . Consistently
with (19.3), we might impose the condition (for large t, at any rate) that
f = e−λt φ at r = 0. (20.2)
Substitution of (20.2) into (20.1) yields

2
∂ φ 4 ∂φ
−λφ = ν + . (20.3)
∂r 2 r ∂r
This equation may be rewritten
a
φ + φ + k 2 φ = 0, (20.4)
r
where
λ
a = 4, k2 = , (20.5)
ν
and primes denote differentiation with respect to r.
We may make the transformation
φ = r n θ. (20.6)
We note that
φ = nr n−1 θ + r n θ
92 20 Discussion of the Case of Negligible Inertial Term
and
φ = n(n − 1)r n−2 θ + 2nr n−1 θ + r n θ ,
so that (20.4) becomes

2n + a n(n − 1 + a)
r θ +
n
θ + + k θ = 0.
2
(20.7)
r r2
If we choose n such that

2n + a = 1,
we find that
1
n = (1 − a). (20.8)
2
With this choice, we conclude that

1−a 1−a
n(n − 1 + a) = +a−1
2 2
2
1−a
=−
2
= −n2 . (20.9)
Hence, for r = 0 at any rate, we find that θ satisfies

1 n2
θ + θ + k 2 − 2 θ = 0, (20.10)
r r
where n is given by (20.8). For a = 4, we have n = −3/2. Thus,
θ = J−3/2 (kr)
is a solution of (20.10). It follows that
θ = J3/2 (kr)
must also be a solution, and since these two Bessel’s functions must be independent,
the complete solution of (20.10) may be written
θ = AJ3/2 (kr) + BJ−3/2 (kr). (20.11)
For r = 0, J−3/2 (kr) is unbounded and so, according to the boundary conditions,
B = 0.
20 Discussion of the Case of Negligible Inertial Term 93
Thus,
A
φ= J3/2 (kr). (20.12)
r 3/2
We may arbitrarily take
c
A= ,
k 3/2
where c is a constant, and may depend on the value of k. In general, k will be an
eigenvalue determined by the boundary conditions. In particular, if we require that
f = 0 at r = r0 , then k will correspond to values for which J3/2 = 0.
We now have
J3/2 (kr)
φ(r) = c . (20.13)
(kr)3/2
Of course,
f (r, 0) = φ(r), (20.14)
and if we require
f (0, 0) = 1,
this is equivalent to requiring
φ(0) = 1. (20.15)
But
∞
(kr)3/2+2km
J3/2 (kr) = (−1)m (20.16)
23/2+2m m!(5/2 + m)
m=0
and
(kr)2m
φ(r) = c (−1)m . (20.17)
23/2+2m m!(5/2 + m)
According to (20.15), we must have
1
c = 1,
23/2 (5/2)
or
c = 23/2 (5/2). (20.18)
The eigenvalues of k will be specified by (19.4). We have
f (r, t) = 23/2 (5/2)J3/2 (kr)e−λt . (20.19)
Clearly,

∂ 2f ∂ 2φ
= ≡ φ (0). (20.20)
∂r 2 0,0 ∂r 2 0
94 20 Discussion of the Case of Negligible Inertial Term
Now
∞
2m(2m − 1)(kr)2m−2 k 2
φ (r) = 23/2 (5/2) (−1)m ,
23/2+2m m!(5/2 + m)
m=1
and
2k 2 k2
φ (0) = − = − . (20.21)
22 · 5/2 5
Then by (19.4),
λ ε
k2 = = , (20.22)
ν 3νu21
from which we find the eigenvalues
ε ε
λ= = . (20.23)
3u21 u2
We note that λ does not depend on the viscosity. Moreover,
du2
ε= ,
dt
and ε/u2 is the life time of an eddy, that is, λ is a relaxation time.
Finally, we note that
1/2
2 sin kr
J3/2 (kr) = − cos kr , (20.24)
πkr kr
and k may thus be determined by the wavelength of the oscillation of f as r → ∞.

However, we found k in this case by evaluating the curvature at the origin.
Chapter 21
The Case in Which Viscosity Is Neglected
If the viscosity is negligible, the equation in f becomes (19.1), viz.,

2
∂ 3f ∂ ∂ 4 ∂
= −f + f. (21.1)
∂t 2 ∂r ∂r ∂r 2 r ∂r
We first attempt to discover the existence of separable solutions. Let
f = φ(t)ψ(r). (21.2)
Then, upon substitution into (21.1), we find

4 4
φ (t)ψ (r) = −φ (t)ψ(r) ψ (r) + ψ (r) − 2 ψ (r) ,
2
r r
where primes denote differentiation with respect to the variable in the argument. We
have then,

φ (t) ψ(r) 4 4
+ ψ (r) + ψ (r) − 2 ψ (r) = 0. (21.3)
φ 2 (t) ψ (r) r r
Then we may set
φ
= −α,
φ2
or
φ + αφ 2 = 0. (21.4)
We multiply (21.4) by φ to obtain
d2 φ dφ dφ
= −αφ 2 , (21.5)
dt 2 dt dt
which may be rewritten
2
1 d dφ α d 3
=− φ . (21.6)
2 dt dt 3 dt
96 21 The Case in Which Viscosity Is Neglected
This may be integrated:

2
dφ 2α 3
=− φ + C1 . (21.7)
dt 3
We know that φ → 0 as t → ∞ and we would like to have dφ/dt → 0 under these
circumstances as well. Thus we are led to set
C1 = 0.
(If we did not set C1 = 0, we would obtain elliptic integrals as solutions.) Further,
we specify that α is negative. Then
dφ
= C2 φ 3/2 , (21.8)
dt
and
dφ
= C2 dt. (21.9)
φ 3/2
We then have
φ −1/2 = C2 t + C3 . (21.10)
C3 will be, in any case, unimportant for large t and we set
C3 = 0.
We thus have
C
φ= , (21.11)
t2
and
C
f= ψ(r). (21.12)
t2
If we put this into (18.11), we find
2
dψ C d ψ 4 dψ
=− d + ,
ψ 6 dr 2 r dr
or

C d2 ψ 4 dψ
+ + ln ψ = C4 . (21.13)
6 dr 2 r dr
If we change scale appropriately, we may write
d2 ψ 4 dψ
2
+ + ln ψ = K. (21.14)
dr r dr
Clearly, (21.14) has the solution
ψ = ψ0 , (21.15)
21 The Case in Which Viscosity Is Neglected 97
Fig. 21.1 Solutions of

(21.14)
where ψ0 is a constant. In particular, ψ0 = eK . We may ask whether other solutions

can be found from the intuitive knowledge that they may approach this solution
asymptotically. Accordingly, we seek solutions of the form
ψ = ψ0 + φ, (21.16)
where
φ → 0 as r → ∞. (21.17)
Upon substitution of (21.16) into (21.14), we have

d2 φ 4 dφ φ
+ + ln ψ 0 1 + = K, (21.18)
dr 2 r dr ψ0
which, since ψ0 = eK , becomes

d2 φ 4 dφ φ
+ + ln 1 + = 0. (21.19)
dr 2 r dr ψ0
Since
∞ n
u
ln(1 + u) = ,
n
n=1
and because of (21.17), we may write approximately, for large r,
d2 φ 4 dφ 1
2
+ + φ = 0. (21.20)
dr r dr ψ0
We saw in the previous section that (21.20) is equivalent to Bessel’s equation and
has the solution
J3/2 (kr)
φ=D , (21.21)
(kr)3/2
where D is a constant and
1
k2 = . (21.22)
ψ0
Thus we see that, for large r, (21.14) admits the solutions shown in Fig. 21.1. We
have therefore seen that, for sufficiently large t and r, the function f is of the form

A J3/2 (kr)
f = 2 ψ0 + B , (21.23)
t (kr)3/2
which is not unlike the form of solution obtained in the pure viscous case.
Chapter 22
Solution of the Non-Viscous Case Near r = 0
We saw that, in the Kolmogorov theory [see (13.20) on p. 57],

2/3
r
f =1− ,
r0
so that, by letting
φ = 1 − f, (22.1)
we find that, for small r, and hence small φ,

∂ 3φ ∂ ∂ 2 φ 4 ∂φ
= − + . (22.2)
∂t 2 ∂r ∂r ∂r 2 r ∂r
Upon integration of (22.2), we have
2
∂ 2φ ∂ φ 4 ∂φ
=− + + h(t), (22.3)
∂t 2 ∂r 2 r ∂r
where h(t) is an arbitrary function of t. The effect of h(t) is to add an arbitrary
function of time to φ in the form of a particular integral of (22.3). We will not be
interested in such an addition and will take
h(t) ≡ 0. (22.4)
Then (22.3) becomes
2
∂ 2φ ∂ φ 4 ∂φ
=− + , (22.5)
∂t 2 ∂r 2 r ∂r
which we must solve with the boundary condition
φ = φ(r, 0), t = 0. (22.6)
It is clear that D5 , the 5-dimensional Laplacian operator, is
5
∂2
D5 = , (22.7)
i=1
∂xi2
100 22 Solution of the Non-Viscous Case Near r = 0
and that (22.5) is
∂ 2φ ∂ 2φ
5
= − . (22.8)
∂t 2 ∂xi2
i=1
We will treat (22.8) as a quasi-wave equation, rather than as Laplace’s equation in 6
dimensions, and will proceed to the solution after an illustrative example in the next
chapter.
Chapter 23
Solution of the Heat Equation
Consider the equation of heat diffusion:

∂θ
= κ∇ 2 θ. (23.1)
∂t
We will set κ = 1/2 and treat
∂θ 1
= ∇ 2 θ. (23.2)
∂t 2
If at t = 0, θ is a unit heat source, the temperature distribution at time t is
1
θ = √ e−x /2t .
2
(23.3)
2t
This may be verified by substitution in (23.2).
A more general problem is: given any arbitrary distribution of θ at t = 0, how is
θ distributed at a later time t? One can express the distribution at t = 0 in terms of
unit sources (δ-functions) in the following way:
+∞
θ (x) = θ (a)δ(x − a)da. (23.4)
−∞
Then given any θ (r) at t = 0, θ (r, t) may be expressed as

+∞

θ (r, t) = θ (r ) |r − r |2 , t dr, (23.5)
−∞
where
A −|r−r |2 /at
= e . (23.6)
t 3/2
Chapter 24
Solution of the Quasi-Wave Equation
In order to make the wave equation invariant under translations of the origin, we
wrote it in the form (22.8):
∂φ 2 ∂ 2φ5
= − . (24.1)
∂t 2 ∂xi2
i=1
The appropriate δ-function (or unit source) is
2 t
. (24.2)
π 3 (t 2 + |r − r |2 )3
To test whether (24.2) is an appropriate δ-function, we merely substitute
2 t
(24.3)
π 3 (t 2 + r 2 )3
into (22.5) and note that it is identically satisfied. In order to test the normalization
of (24.2), we investigate the integral
+∞

2 dV
t , (24.4)
π3 (t 2 + r 2 )3
−∞
where dV = dx1 dx2 dx3 dx4 dx5 , or in terms of spherical polars:

⎫
x1 = r cos θ, ⎪
⎪
⎪
⎪
⎪
⎪
x2 = r sin θ cos φ1 , ⎪
⎪
⎬
x3 = r sin θ sin φ1 cos φ2 , (24.5)
⎪
⎪
⎪
x4 = r sin θ sin φ1 sin φ2 cos φ3 , ⎪
⎪
⎪
⎪
⎪
⎭
x5 = r sin θ sin φ1 sin φ2 sin φ3 .
104 24 Solution of the Quasi-Wave Equation
Fig. 24.1 Transformation for

calculating the integral in
(24.7)
The integral (24.4) becomes

∞ π
2 π π 2π r 4 sin3 θ sin2 φ1 sin φ2 dθ dφ1 dφ2 dφ3 dr
t
π3 0 0 0 0 0 (t 2 + r 2 )3
∞
16 r 4 dr
= t . (24.6)
3π 0 (t 2 + r 2 )3
Putting r = ty, we find
∞ ∞
16 r 4 dr 16 y 4 dy
t = . (24.7)
3π 0 (t + r )
2 2 3 3π 0 (y 2 + 1)3
If we perform the transformation implied by Fig. 24.1, we find that (24.7) is
∞
16 y 4 dy 16 π/2
= sin4 α dα = 1. (24.8)
3π 0 (y + 1)
2 3 3π 0
Note that it may also be shown that (24.2) is a unique point source in this problem.
As in the last chapter, if we are given an initial distribution
(r) at t = 0,
we have at time t
+∞

2 (r )dx1 dx2 dx3 dx4 dx5
φ(r, t) = t . (24.9)
π3 (t 2 + |r − r |2 )3
−∞
Equation (24.9) is the solution of the boundary value problem of (22.8). For the
spherically symmetric case
∞ π π π 2π
2t (r )r 4 sin3 θ sin2 φ1 sin φ2 dr dφ1 dφ2 dφ2 dθ
φ(r, t) =
π3 0 0 0 0 0 (t 2 + r 2 + r 2 − 2rr cos θ )3
∞
4t π (r )r 4 dr
= sin3 θdθ . (24.10)
π 0 0 (t 2 + r 2 + r 2 − 2rr cos θ )3
This provides a unique solution. The integration over θ is messy and is not per-
formed here.
24 Solution of the Quasi-Wave Equation 105
In the framework of isotropy, the integration over infinite r is not permissible. On

the other hand, if the integral converges, no large contribution will be made at large
r and so this integral is approximately correct, even in isotropy, if it converges. We
assume convergence, and thus have a restriction on (r ), namely
(r ) r . (24.11)
In particular, (r ) ∼ (r )2/3 is permitted by this restriction.

If in (24.10), we replace r by the complex variable z, the integrand has no zeros
on the real axis, but its denominator has zeros in the complex plane. The calculus of
residues can therefore be used in evaluation of the integral, if we are careful of the
difficulty introduced by the fact that the origin is a branch point.
The complex integral is
∞
4t π z4+a dz
φ= dθ sin3 θ . (24.12)
π 0 0 (t 2 + r2 + z2 − 2rz cos θ )3
This becomes

(4 + a)(3 + a) a π−arctan x
φ = ra x(x 2 + 1) (1 − x 2 ) cot2 ψ sin(2 + a)ψdψ,
2 sin aπ 2 arctan x
(24.13)
where
t
x= . (24.14)
r
We see, therefore, that
φ = r a ψ(x). (24.15)
Clearly, φ as given by (24.15) should satisfy (22.8) and we may substitute (24.15)
back in. This procedure should give us an equation in ψ(x).
We note that
dx t X
=− 2 =− . (24.16)
dr r r
In making the substitution into (24.15), we find it necessary to make the following
calculations:
∂φ
= r a−1 ψ (x), (24.17)
∂t
∂ 2φ
= r a−2 ψ (x), (24.18)
∂t 2
∂φ
= ar a−1 ψ(x) − r a−1 xψ (x)
∂r

= r a−1 aψ(x) − xψ (x) , (24.19)
∂ 2φ
2
= (a − 1)r a−2 aψ(x) − xψ (x) − r a−2 x aψ (x) − ψ (x) − xψ (x)
∂r

= r a−2 x 2 ψ − 2(a − 1)xψ + a(a − 1)ψ , (24.20)
4 ∂φ

= 4r a−2 −xψ + aψ . (24.21)
r ∂r
Then (22.5) becomes

ψ (x) = − x 2 ψ (x) − 2(a + 1)xψ (x) + a(a + 3)ψ(x) , (24.22)
or
(1 + x 2 )ψ − 2(a + 1)xψ + a(a + 3)ψ = 0. (24.23)
Kamke’s dictionary of differential equations gives
(ax 2 + 1)y + bxy + cy = 0, (24.24)
and adds that (24.24) admits solutions in closed form if
(a − b)2 − 4ac = (2n + 1)2 a 2 ,
where n is an integer. In our case,
a = 1, b = −2(a + 1), c = a(a + 3),
and so
2
1 + 2(a + 1) − 4a(a + 3) = (2n + 1)2 ,
(2a + 3)2 − 4a(a + 3) = (2n + 1)2 ,
9 = (2n + 1)2 ,
from which n = 1. Hence, (24.24) may be solved explicitly, and Kamke indicates
that Forsythe gives the theory of this type of equation.
To solve (24.23), we let
x = eτ . (24.25)
Then
dτ
= e−τ , (24.26)
dx
∂ψ ∂ψ ∂τ ∂ψ −τ
= = e , (24.27)
∂x ∂τ ∂x ∂τ
2
∂ 2ψ ∂ ∂ψ −τ −τ ∂ ψ ∂ψ −2τ
= e e = − e . (24.28)
∂x 2 ∂τ ∂τ ∂τ 2 ∂τ
Let
∂n
Dn = . (24.29)
∂τ n
Then (24.23) becomes

1 − e−2τ (D 2 − D)ψ − 2(α + 1)Dψ + α(α + 3)ψ = 0. (24.30)
Then
e−2τ (D 2 − D)ψ + D 2 ψ − (2α + 3)Dψ + α(α + 3)ψ = 0,

D(D − 1)ψ + e2τ D 2 − (2α + 3)D + α(α + 3) ψ = 0, (24.31)
D(D − 1)ψ + e2τ (D − α)(D − α − 3)ψ = 0.
Now
φ(D)enτ f = enτ φ(D + n)f, (24.32)
so (24.31) can be written
D(D − 1)ψ + (D − α − 2)(D − α − 5)e2τ ψ = 0, (24.33)

(D − α − 2)(D − α − 5) 2τ
ψ+ e ψ = 0. (24.34)
D(D − 1)
Let
ψ = (D − α − 2)ψ1 . (24.35)
Then
(D − α − 5) 2τ
ψ1 + e (D − α − 2)ψ1 = 0, (24.36)
D(D − 1)
and
(D − α − 5)(D − α − 4) 2τ
ψ1 + e ψ1 = 0, (24.37)
D(D − 1)
D−α−4 τ D−α−4 τ
ψ1 + e e ψ1 = 0,
D D

D−α−4 τ 2
1+ e ψ1 = 0, (24.38)
D
D−α−4 τ
±iψ1 + e ψ1 = 0,
D
±iDψ1 + eτ (D − α − 3)ψ1 = 0,

τ dψ1
e ±i − (α + 3)eτ ψ1 = 0,
dτ

α+3
ψ1 = A eτ ± i , (24.39)

α+3
α+2
ψ = A eτ ± i ∓ (α + 3) eτ ± i , (24.40)
or

α+3
α+2
ψ =A x ±i ∓ i(α + 3) x ± i . (24.41)
Now,

1
x ± i = (x + 1)
2 1/2
exp ±i arctan . (24.42)
x
Setting
θ = arctan(1/x), (24.43)
we have
x ± i = (x 2 + 1)1/2 e±iθ ,
(x ± i)α+3 = (x 2 + 1)(α+3)/2 e±i(α+3)θ

= (x 2 + 1)(α+3)/2 cos(α + 3)θ ± i sin(α + 3)θ .
Similarly, the second terms become

∓i(x ± i)α+2 = ∓i(x 2 + 1)(α+2)/2 cos(α + 2)θ ± i sin(α + 2)θ

= (x 2 + 1)(α+2)/2 sin(α + 2)θ ∓ i cos(α + 2)θ ,
and (24.41) becomes

ψ = A (x 2 + 1)(α+2)/2 (x 2 + 1)1/2 cos(α + 3)θ + (α + 3) sin(α + 2)θ

± i(x 2 + 1)(α+2)/2 (x 2 + 1)1/2 sin(α + 3)θ − (α + 3) cos(α + 2)θ .
(24.44)
We may take as the linearly independent solutions the real and imaginary parts of
(21.7). Thus the solutions are
⎫
ψ1 = (x 2 + 1)(α+2)/2 (x 2 + 1)1/2 cos(α + 3)θ + (α + 3) sin(α + 2)θ , ⎪
⎬

⎪
ψ2 = (x 2 + 1)(α+2)/2 (x 2 + 1)1/2 sin(α + 3)θ − (α + 3) cos(α + 2)θ . ⎭
(24.45)
Now the general solution of (24.23) will be some linear combination of ψ1 and
ψ2 , as given by boundary conditions. We note that:
• for r = 0, as t → 0, x → 0 and θ → π/2,

• for t = 0, as r → 0, x → ∞ and θ → 0.
Hence, in the neighbourhood of r = 0, we may note that
ψ1 → x α+3 as x → ∞,
and
a+3
t t a+3
g1 → r a → 3 → ∞.
r r
Chapter 25
The Introduction of Boundary Conditions
We had

∂ ∂2 ∂
− ν D5 f = f D5 f.
2 2
(25.1)
∂r ∂t 2 ∂r
In applying the boundary conditions, we saw that going to r → ∞ is incompatible
with local isotropy and we were restricted to r < r0 . Moreover, if τ is the lifetime of
the largest eddies, we are similarly restricted to t < τ . Thus the condition f → 0 as
t → ∞ cannot be applied. The lack of boundedness of f as r and t tend to infinity
is the result of the ordering of the correlation imposed by the largest eddies. Another
condition must therefore be sought.
One thing appears now as strange: the Kolmogorov principles as applied to the
velocity correlations are incompatible with the new theory. Kolmogorov applies his
principles to the relative velocity correlations, i.e., to
1
ψ= u − u1 )2 . (25.2)
2 1
If we now choose to measure quantities in the dimensionless units, we must have,
according to the first principle,

ψ = (εν)1/2 r(ε/ν 3 )1/4 . (25.3)
Moreover, if the second principle is applied, we find that
(x) → Cx 2/3 as ν → 0. (25.4)
Now
ψ = u21 − f (r, t), (25.5)
and if we put this into (25.1), there results
2
∂ ∂ ∂ ∂
− ν 2 2
D 5 ψ = −ψ D5 ψ + u21 D5 ψ. (25.6)
∂r ∂t 2 ∂r ∂r
112 25 The Introduction of Boundary Conditions
The ordinary application of Kolmogorov’s principle must now be extended to the

present case of correlation over a time interval. We must choose a suitable time unit.
An appropriate choice can be found by recalling that
L2 L2
[ε] = , [ν] = . (25.7)
T3 T
Hence we may express ψ as

ψ = (εν)1/2 r(ε/ν 3 )1/4 , t (ε/ν)1/2 , (25.8)
and if we substitute the foregoing into (25.6), we get an appropriate equation for .
However, the appearance of u21 is in violation of the first Kolmogorov principle.
One therefore feels that either the theory or the application of the Kolmogorov
principles to ψ is incorrect. The latter supposition seems the more reasonable. This
is suggested by the following argument. From the Kolmogorov principles, we found
that
ψ → const. × (εr)2/3 ,
and hence is unbounded. But
(u1 − u1 )2
must be bounded, so that the application of Kolmogorov’s principles to ψ leads to a
contradiction. It is therefore natural to seek some other physical variable which may
be more suited to the application of the principles.
Consider
∂ψ
χ =− .
∂r
This quantity goes as r −1/3 and so is bounded even in the Kolmogorov theory. More-
over, it contains no reference to r0 and u1 , which is the aspect of ψ that leads to dif-
ficulty. The lack of reference to r0 and u1 removes the necessity of having to choose

between the two possible fundamental length scales, r0 and the velocity unit u21 .
We may now see what Chandrasekhar’s equation is for x. We write (25.6) as
∂ ∂2
∂r ( ∂t 2 − ν D5 )ψ
2 2
∂
= u21 − ψ.
∂r D5 ψ
Then
⎡ ⎤
∂ ∂2
∂ ⎣ ∂r ( ∂t 2 − ν D5 )ψ ⎦
2 2
∂
= χ,
∂r ∂r D5 ψ
and we see that the extra term due to the u21 is lost in the differentiation. Another
aspect in looking at the behavior of χ is to discover its form in the Kolmogorov
25 The Introduction of Boundary Conditions 113
theory. If we merely differentiate (25.8), we find that
∂(x, y) ∂x
χ = (εν)1/2 ,
∂x ∂r
where x = (ε/ν 3 )1/4 r, y = (ε/ν)1/2 t, and clearly ∂y/∂r = 0. Thus,
1/4 1/4
ε3 ∂ ε3
χ= = X,
ν ∂x ν
where X, the derivative of a universal function, is itself a universal function. In order

to facilitate the transition to ν → 0, we choose as the unit of time r 2/3 /ε 1/3 . Then
3 1/4 3 1/4
ε ε t 1/3
χ= X r, 2/3 ε .
ν ν r
Now, as ν → 0, one argument of X goes to infinity. But if we assume the power law
and the second principle of Kolmogorov, we require that as ν → 0,
1/3
tε1/3 tε
X x, 2/3 → x −1/3 σ .
r r 2/3
Also, χ → 0 as r → ∞, or t → ∞.
There is some physical basis for the choice of χ as a variable. We have
∞
J3/2 (kr)
f= F (k) dk.
0 (kr)3/2
But in local isotropy, F (k) → k −5/3 , so that the integral diverges. But on differenti-
ating the foregoing, we obtain (see Chamberlain and Roberts)
1/2 ∞
1 ∂ 3 2
(r D5 f ) = − kF (k) sin kr dk,
r ∂r π 0
which converges. The way to make f converge would be to cut off the integration
at some k0 , but this would imply a reference to the largest eddies. D5 f is essen-
tially the defining scalar of the vorticity. This is a local property whose description
requires no reference to the larger eddies, i.e., to r = 0.
Postscript Here the notes end with a remark that points to the difference of the
results of the statistical theories of the era from those of Kolmogorov’s theory of
1941. The data at the time did not yet completely confirm K41 but it was the ac-
cepted version of things and the problem was to bring conformity with it.
Epilogue
These notes do not end with a bang, for not only was the summer ending but Chan-
dra was still wrestling with some technical details. I returned to the Yerkes Obser-
vatory for a two-week visit in the summer of 1955 and Chandra gave a seminar on
the then current state of the theory. He was getting results in conflict with those of
Kolmogorov and he had concluded that “either the theory or the application of Kol-
mogorov’s principles to was inappropriate.” As one knows nowadays, a problem
arises from the use of Eulerian coordinates in this context.
Still, Chandra pulled things together and published two papers on his approach
(in 1955 and 1956). The initial reception of the theory was positive. Indeed, Stanley
Corrsin once told me that, back in the mid-fifties, he was so sure that the “turbu-
lence problem” would soon be solved that he bet George Uhlenbeck five dollars that
he was right. Afterwards, when Corrsin and Uhlenbeck heard Chandra lecture on
his theory, Uhlenbeck came over and handed Corrsin a fiver. It soon appeared that
Uhlenbeck should have waited before parting with his money.
Numerical treatments of the theory revealed that the energy spectrum goes neg-
ative for some wavenumbers. Kraichnan pointed out that all such theories, which
he called quasi-normal, and especially the multi-time ones such as Chandra’s and
an earlier one by Heisenberg, contained inconsistencies. Of course, the hope of the
quasi-normal theory was that, by making approximation on the fourth-order corre-
lation, one does not do too much damage to the energy conserving property of the
triple correlation. The problem is that the danger is not merely quantitative. Early
in these notes, Chandra remarks that phase relations are important in discussing the
statistics of turbulence. In his 1955 paper he wrote:
A description in terms of F (k) only (or Q(r) only) would be complete only if there were no
phase relationships between the different Fourier components of the velocity field. But this
is not the case. Phase relationships must exist: without them there would be no exchange
of energy between the different Fourier components which is, after all, the essence of the
phenomenon of turbulence. A theory, albeit an approximate theory, must incorporate in
itself some element which describes these phase relationships; without such an element
the theory would lack the means of accounting for the essence of the phenomenon. It would
appear that by introducing the correlations in the velocity components at two different points
and at two different times, we can incorporate features which are the result of these phase
relationships.
DOI 10.1007/978-94-007-0117-5, © Springer Science+Business Media B.V. 2011
116 Epilogue
His attempt to come to grips with this aspect of the statistics of turbulence turned
out to be a significant source of trouble in his theory. The phase relations cannot be
dealt with so easily, though the thought seems a good one. Unfortunately, it is in the
multitime formulations that the inconsistencies in the quasi-normal theory show up
most vivdly as Kraichnan elucidated in a paper of 1959. That matter is clarified also
in the thesis1 that Gary Deem wrote under J.B. Keller’s supervision.2
But Chandra was not alone in finding that problems arose in applying the quasi-
normal theories in Eulerian coordinates. In May 1955, I heard a lecture by George
Batchelor in which he reported difficulties encountered by Proudman and Reid in
their version of the approach in the form of a singularity at small wave number.
Batchelor explained the problem in terms of the nonlocal behavior caused by the
pressure gradient, though he made no mention of the role of incompressibility in
such difficulties.
It is surprising that Chandra made no mention of any other quasi-normal theories
than his own, though several existed by the time of his lectures, going back to Mil-
liontschikov in 1941 (the same year that Kolmogorov’s paper appeared). For better
or worse, I have decided to keep to the informal style of the original notes from
Chandra’s lectures and to not include a set of relevant references, whose informa-
tion is easily obtained these days. But indeed others had developed quasi-normal ap-
proximations for the statistical theory of turbulence and, in a paper on these matters,
Kraichnan, gave a fairly complete list of them3 along with an extensive discussion
of the issues they present.
In his thesis, Deem tried to express the distinction among the quasi-normal the-
ories in terms of the different time integration paths taken to develop them in the
space of the different times involved. I have wondered whether something like
Caratheodory’s approach to thermodynamics might be introduced to enhance the
understanding that Deem’s work suggests, but this can be effective in only certain
special cases. A much more direct way to modify Chandra’s quasi-normal theory
into something usable (called the EDQNM model) was given by Orszag and pur-
sued by the group (GRETPA)4 in the Nice Observatory. Perhaps the real problem
was that none of these statistically motivated approaches made attempts to bring
their theories into conformity with the physical imagery behind Kolmogorov’s the-
ory, a point well made by Kraichnan. Now there are books discussing all that and
this is not the place to write another one. Rather, with Chandra’s centennial upon us,
it is more appropriate to make Chandra’s notes available. Their value is that, besides
being a very simple introduction to the theory of homogeneous turbulence, they
show us Chandra’s approach to science. He was, above all, a fine stylist who lent
elegance and clarity to the subjects he treated through purely mathematical efforts.
1 Kraichan,R.H.: The structure of isotopic turbulence at very high Reynolds number, Journal
of Fluid Mechanics, Vol. 5, 1959, p. 497.
2A Nearly Normal Theory for Decaying Zero-Mean Turbulence, Thesis, NYU, 1969.
3 Inthe Proceedings of Symposia in Applied Math. 13 “Hydrodynamic Instability” (Am. Math.
Soc., 1962), pp. 199–225.
4 GRETPA: Groupe de Recherche sur la Turbulence et les Phénomènes Aléatoires.
Epilogue 117
I believe that his lectures had a twofold purpose. They not only provided a very
elementary introduction to some aspects of the subject for novices, they also allowed
Chandra to organize his thoughts in preparation to formulating his attack on the
statistical problem of homogeneous turbulence. He told me, in that summer of 1954,
that he continually said to himself that he must find a way to close the equations for
the moment hierarchy. From the notes, we get some flavor of how Chandra worked.
His starting thoughts in a subject were not so different from the final forms you find
in his books. Above all, he let the mathematics be his guide. This kind of approach
seems to be in the spirit advocated by Dirac, though I believe that Dirac kept the
physical details more firmly in mind and had the good sense to stick mainly to
linear problems.
But, joking aside, I find it intriguing that the ideas that have most influenced
the course of turbulence theory in the past fifty years were the thoughts of another
mathematician. Komogorov’s calculation may seem to be simplicity itself, but it is
based on a powerful vision of the physics that continues to inspire the mathematical
thinking about turbulence to this day. And in preparing these notes I see more clearly
the truth in Kraichnan’s criticism that the proposers of the seemingly harmless quasi-
normal theories did not make much effort to bring their statistical approaches into
line with Kolmogorov’s vision. In fact, as Heisenberg noted, this could not be done
in Eulerian coordinates.
Thus did Chandra join the list of distinguished physical scientists who had tilted
at the turbulence problem and managed only to uncover further difficulties in its
analysis. It is interesting that one does not find the name of Einstein on that list,
though he turned out to be a grey eminence in the subject. Kraichnan was (what we
now call) a postdoc at the IAS in Princeton, where he was associated with Einstein,
who advised him to go into fluid dynamics. Thus did Kraichnan go on to clarify
the inconsistencies of the quasi-normal theories and, by his own contribution, to
leave the imprint of his quantum field theoretical background on current theories of
turbulence.

The Theory of Turbulence - Subrahmanyan Chandrasekhar's - Spiegel E.A. (Ed.)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Theory of Turbulence - Subrahmanyan Chandrasekhar's - Spiegel E.A. (Ed.)

Uploaded by

Copyright:

Available Formats

Lecture Notes in Physics

Founding Editors: W. Beiglböck, J. Ehlers, K. Hepp, H. Weidenmüller

For other titles published in this series, go to

Notes prepared by E.A. Spiegel

Spiegel, E.A., The Theory of Turbulence: Subrahmanyan Chandrasekhar’s 1954

ISSN 0075-8450 e-ISSN 1616-6361

Library of Congress Control Number: 2010938378

© Springer Science+Business Media B.V. 2011

Cover illustration: eStudio Calamar S.L.

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

At a Midwest Neighborhood Astronomy Meeting in the spring of 1953, I gave a

S. Chandrasekhar (1910–1995), originally published in 1952 in the publications

A cascade. Photograph by Antonello Provenzale

1 The Turbulence Problem . . . . . . . . . . . . . . . . . . . . . . . . . 1

2 The Net Energy Balance . . . . . . . . . . . . . . . . . . . . . . . . . 3

3 Interchange of Energy Between States of Motion . . . . . . . . . . . 7

5 The Spectrum of Turbulent Energy . . . . . . . . . . . . . . . . . . . 13

6 Some Preliminaries to the Development of a Theory

7 Heisenberg’s Theory of Turbulence . . . . . . . . . . . . . . . . . . . 21

8 Other Derivations of the k −5/3 Law . . . . . . . . . . . . . . . . . . . 29

9 An Alternate Approach: Correlations . . . . . . . . . . . . . . . . . 35

10 The Equations of Isotropic Turbulence . . . . . . . . . . . . . . . . . 37

11 The Karman-Howarth Equations . . . . . . . . . . . . . . . . . . . . 49

12 The Meanings of the Defining Scalars . . . . . . . . . . . . . . . . . . 51

13 Some Results from the Karman–Howarth Equation . . . . . . . . . . 55

14 The Relation Between the Fourth and Second Order Correlations

15 Chandrasekhar’s Theory of Turbulence . . . . . . . . . . . . . . . . 71

16 A More Subjective Approach to the Derivation of Chandrasekhar’s

17 The Dimensionless Form of Chandrasekhar’s Equation . . . . . . . 83

18 Some Aspects and Advantages of the New Theory . . . . . . . . . . . 85

19 The Problem of Introducing the Boundary Conditions . . . . . . . . 89

20 Discussion of the Case of Negligible Inertial Term . . . . . . . . . . . 91

21 The Case in Which Viscosity Is Neglected . . . . . . . . . . . . . . . 95

22 Solution of the Non-Viscous Case Near r = 0 . . . . . . . . . . . . . . 99

23 Solution of the Heat Equation . . . . . . . . . . . . . . . . . . . . . . 101

24 Solution of the Quasi-Wave Equation . . . . . . . . . . . . . . . . . . 103

25 The Introduction of Boundary Conditions . . . . . . . . . . . . . . . 111

1.1 The Meaning of ‘Turbulence’

1.2 Two Fundamental Aspects of Turbulence

(a) Viscous Dissipation of Energy. If energy is supplied to a system at a fixed rate,

where dS is a vector normal to the surface and of arbitrary magnitude. If we multiply

(a) The First Term on the Left-Hand Side:

where T is the kinetic energy contained in the volume.

since uj is solenoidal. Then, by the divergence theorem,

(c) The Pressure Term:

ui ∇ 2 ui = −|curl u|2 + div (u × curl u),

proved at the end of this section. Then, in view of this identity,

Let ω = curl u. Then, on gathering the various terms of (2.2) we obtain

Proof of the Lemma

Combining (2.4) and (2.5) leads to

|curl u|2 = (δj m δkn − δj n δkm )uk,j un,m

(u × curl u)i = ij k uj kmn un,m

= 4πlA (R22 − R12 ),

Here k is the wave vector, corresponding to the wavelength

Equation (3.3) will be true in general (only) if

The orthogonality of uk and k implies that the time variation of uk is a rotation in

Another convenient step is to rewrite (1.1) in vector form. We then have

Finally, under application of (3.4), the term becomes

The resolved form of (3.7) in final form is thus

If the eik·r coefficients are equated term by term, there results

4.1 On the Harmonic Analysis

We have heretofore considered finite volumes over which to Fourier-analyze veloc-

(u × curl u)i = ij k uj kmn un,m

where ij (k) is a generalized spectrum.